A Survey of The State-of-the-Art Localization Techniques and Their Potentials For Autonomous Vehicle Applications

IEEE INTERNET OF THINGS JOURNAL, VOL. 5, NO.
2, APRIL 2018 829
A Survey of the State-of-the-Art Localization

Techniques and Their Potentials for
Autonomous Vehicle Applications
Sampo Kuutti, Saber Fallah , Konstantinos Katsaros, Mehrdad Dianati,
Francis Mccullough, and Alexandros Mouzakitis
Abstract—For an autonomous vehicle to operate safely and considered as a potential solution to some of these problems by
effectively, an accurate and robust localization system is essen- providing more efficient driving and proper vehicle control [2].
tial. While there are a variety of vehicle localization techniques The architecture through which autonomous driving is
in literature, there is a lack of effort in comparing these
techniques and identifying their potentials and limitations for achieved can be described by five functional systems, includ-
autonomous vehicle applications. Hence, this paper evaluates the ing localization, perception, planning, control, and system
state-of-the-art vehicle localization techniques and investigates management [3]. The localization system identifies the loca-
their applicability on autonomous vehicles. The analysis starts tion of the vehicle on a global coordinate system while the
with discussing the techniques which merely use the informa- perception system evaluates the driving environment around
tion obtained from on-board vehicle sensors. It is shown that
although some techniques can achieve the accuracy required for
the vehicle and identifies elements such as other road users,
autonomous driving but suffer from the high cost of the sen- traffic signals, and obstacles. The planning system uses the
sors and also sensor performance limitations in different driving inputs from the perception and localization systems to deter-
scenarios (e.g., cornering and intersections) and different environ- mine the travel paths and driving actions such as lane changes,
mental conditions (e.g., darkness and snow). This paper continues accelerating, and braking. The control system transforms the
the analysis with considering the techniques which benefit from
behavior and actions indicated by the planning system to the
off-board information obtained from V2X communication chan-
nels, in addition to vehicle sensory information. The analysis vehicle manipulation commands (e.g., steering, accelerating,
shows that augmenting off-board information to sensory informa- and braking). The system management oversees the operation
tion has potential to design low-cost localization systems with high of all the systems and provides the human–machine interface.
accuracy and robustness, however, their performance depends on These functional systems must cooperate to answer the three
penetration rate of nearby connected vehicles or infrastructure important questions of “Where is the car?” “What is around
and the quality of network service.
the car?” and “What does the car need to do next?” to achieve
Index Terms—Autonomous vehicles, cooperative systems, fully autonomous operation.
global positioning system (GPS), intelligent vehicles, Internet The perception, planning, and control functional systems
of Things (IoT), sensor fusion, simultaneous localization and
mapping, vehicular ad hoc networks. require accurate knowledge of the vehicle’s location to carry
out the correct driving decisions and actions. For example, an
error of few decimeters can cause the vehicle to localize itself
I. I NTRODUCTION on the wrong side of the road or can cause accidents to the vul-
ETWEEN 2000 and 2014 the number of registered road nerable users such as pedestrians and cyclists. Robustness is
B vehicles in the U.S. has increased by 35 million [1], an
increase of 15%. This rapid increasing number of road vehicles
also required as the vehicle needs to localize itself in uncer-
tain driving conditions (e.g., in cases where road markings
has amplified the importance of issues such as traffic conges- are absent or obscured) and in harsh environmental conditions
tion, pollution, and road accidents. Autonomous vehicles are (e.g., darkness and snow). Therefore, autonomous vehicles
require robust localization systems with centimeter level of
Manuscript received August 30, 2017; revised December 2, 2017 and
February 13, 2018; accepted February 18, 2018. Date of publication
accuracy [4]. The global positioning system (GPS) is the most
March 5, 2018; date of current version April 10, 2018. This work was sup- commonly used localization system for vehicle applications,
ported by Jaguar Land Rover and the U.K.-EPSRC as part of the jointly funded as they offer a cheap and easily accessible solution for global
Towards Autonomy: Smart and Connected Control (TASCC) Programme positioning. However, GPS suffers from poor reliability due
under Grant EP/N01300X/1. (Corresponding author: Saber Fallah.)
S. Kuutti and S. Fallah are with the Centre for Automotive to multiple limitations such as signal blockage and multipath
Engineering, University of Surrey, Guildford GU2 7XH, U.K. (e-mail: as well as inadequate accuracy (∼10 m) for autonomous
s.j.kuutti@surrey.ac.uk; s.fallah@surrey.ac.uk). vehicles. In order to develop robust, accurate, and reliable
K. Katsaros is with the Technology Department, Digital Catapult, London
NW1 2RA, U.K. (e-mail: kostas.katsaros@digicatapult.org.uk). localization systems for autonomous vehicle applications,
M. Dianati is with the Warwick Manufacturing Group, University of recent research has focused on the development of localiza-
Warwick, Coventry CV4 7AL, U.K. (e-mail: m.dianati@warwick.ac.uk). tion systems either using advanced sensors (e.g., RADAR,
F. Mccullough and A. Mouzakitis are with Jaguar Land Rover Ltd.,
Coventry CV4 7HS, U.K. (e-mail: fmccull2@jaguarlandrover.com). LiDAR, camera, etc.) or fusing on-board and off-board
Digital Object Identifier 10.1109/JIOT.2018.2812300 information.
2327-4662 c 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
830 IEEE INTERNET OF THINGS JOURNAL, VOL. 5, NO. 2, APRIL 2018
Advanced sensors can offer more accurate and reliable posi- types used for vehicle localization is required. This section
tion measurements than GPS, but at an increased cost. These defines types of maps relevant to vehicle localization which
sensors typically operate by sensing information about the are exploited by localization techniques described in the
environment, such as road marks, obstacles, other road users following sections. In general, there are two main categories
etc. However, their performance is limited due to direct line of of maps: 1) planar which refers to maps that rely on layers
sight requirement and lack of robustness against harsh envi- or planes on a geographic information system (GIS), e.g.,
ronmental situations (e.g., snow, darkness, etc.). Therefore, high-definition (HD) maps and 2) point-cloud which refers to
any obstacle which is obscured cannot be identified, thereby maps based on a set of data points in the GIS. The generation
degrading the real-time system performance. Moreover, the of such maps from companies such as HERE, Google, Uber,
cost-effectiveness is a further drawback of these advanced sen- and TomTom, is challenging and pertains initial data collec-
sors, in order to satisfy the high accuracy requirements of tion, data analysis, and continuous maintenance. Such maps
autonomous vehicles. However, with the emergence of Internet can be maintained exploiting near-real time crowdsourcing
of Things (IoT) [5], it is possible for a vehicle to access to information from vehicles traveling and updating backend
a wider range of information (e.g., states of nearby vehicles, map systems, e.g., Mobileye’s road experience management
weather information, traffic status, etc.). By connecting vehi- system [11]. The main difference in using these two types
cles to intervehicle network, intravehicle network, and vehic- of maps is the type of sensor responsible for localizing
ular mobile internet, an IoT infrastructure application can be the vehicle. Vision-based sensors such as cameras, light
formed which is referred to as Internet of Vehicles (IoV) [6]. detection and Ranging (LiDAR), radio detection and rang-
Using IoV, it is possible to enable the next generation of intel- ing (RADAR), and ultrasonic use primarily point-cloud maps,
ligent transportation systems [7]–[9]. Providing wireless con- whereas GPS-based systems use planar maps. Map matching,
nectivity to vehicles enables communication with internal and i.e., the process of localizing the vehicle with respect to the
external environments, supporting vehicle-to-vehicle (V2V), underlying map, is more complex in point cloud systems as
vehicle-to-infrastructure (V2I), vehicle-to-sensor, and vehicle- it requires intense computations, both at the sensor itself (see
to-Internet communications [10]. Such information can be the next section for details) and for the identification of the
integrated to localization systems to further improve both accu- points. For that reason, systems such as the NVIDIA DRIVE
racy and robustness and to address the line of sight issues PX2 [12] introduce high computation capabilities on-board
associated with on-board sensors at a relatively low cost. The the vehicle. Nevertheless, there is no clear advantage of one
localization systems which benefit from off-board information type over the other in terms of accuracy of the localization as
are known as cooperative localization systems. It is noted that the generation of HD/3-D-maps entail usage of vision-based
in the context of cooperative localization systems, off-board sensors during the collection phase.
information can be received either from nearby vehicles, fixed Planar maps are generated by capturing and analyzing
roadside infrastructure units, or both. data from high-resolution satellite or aerial photography, GPS
There are various research works in the literature proposing traces and imaging. The process of detailed recording the
different localization techniques with different approaches in environment for later modeling is called micro-mapping in
sensor configuration, data fusion methodologies, connectivity, general [13]. The resolution of those maps depends on the
and feature map creation. However, there is a lack of analyz- captured data and enrichment with additional layers of infor-
ing and comparison of these different approaches to identify mation from analyzed data. For example, HD maps provide
their potentials and limitations which can be highly benefi- a base map layer enriched with a precise sublane level
cial toward deployment of autonomous cars to the market. representation of the road network, including the lane mark-
Therefore, this report aims to survey the state-of-the-art local- ings, road furniture, and curvature. With additional details
ization techniques and analyze their performance. To this end, from stereoscopic cameras, 3-D maps provide also the height
the state of the current field is evaluated and recommendations of objects on the map. In addition to static information
for future research in vehicle localization systems are made. about the environment topology, dynamic information such as
The remainder of this paper is structured as follows. the traffic conditions, events on the road, location of other
Section II describes mapping techniques used in vehicle local- moving objects can be incorporated into maps. A standard
ization context. Section III discusses recent localization tech- method to represent such information is the local dynamic
niques, which utilize only on-board sensors while Section IV map (LDM) [14], [15], that contains several layers of infor-
is devoted to analyzing cooperative localization techniques. mation. Such dynamic information can be used to self-adapt
Finally, Section V concludes the current state of the localiza- localization and makes maintenance of accurate map more
tion techniques and technologies and makes recommendation efficient. For instance, Fig. 1 illustrates different layers of
for the direction of future work in the field. an LDM which can be used for the localization systems of
an autonomous vehicle. The first layer represents the static
elements of a map, for example, the road network, build-
II. M APPING T ECHNIQUES ings, and other permanent features of a map. This information
Localization is in principle the process of positioning an is usually provided by a map supplier. The second layer
object, in the context of this report a vehicle, a pedestrian contains quasi-static features such as road-side infrastructure,
or other road furniture, with respect to a reference global or variable speed limits. These are obtained during the opera-
local map. In this sense, a proper definition of alternative map tion of the system. The third layer contains transient dynamic
KUUTTI et al.: SURVEY OF STATE-OF-THE-ART LOCALIZATION TECHNIQUES AND THEIR POTENTIALS 831
Fig. 1. Local dynamic map layers with example information. Adapted

from [17].
Fig. 3. Differential GPS correction. Gray objects indicate estimated position

from uncorrected GPS data, while black figures indicate the ground truth. The
base station calculates the GPS error from its own GPS measurements and
known ground truth, which is then transmitted to the vehicle.
coordinate system such as Earth-centered inertial (ECI) coor-

dinate system, Earth-centered Earth-fixed (ECEF) coordinate
Fig. 2. Point cloud captured from LiDAR on-board a vehicle [18]. system, or the geographical coordinate system. For the brevity
of this paper, the detailed definition of each coordinate system
data such as weather information, traffic light timings, and is referred to [19]. The main sensors considered in this section
phase and congestion levels, obtained through interaction include GPS, inertial motion units (IMUs), cameras, RADAR,
with the infrastructure. Finally, the fourth layer of an LDM LiDAR, and ultrasonic sensors. The following sections provide
contains highly dynamic data through vehicle-to-everything details of the capabilities of each sensor including benefits and
(V2X) communications and contains location information of limitations as well as analysis of localization techniques using
other vehicles, road-side units (RSUs), even pedestrians or each sensor standalone or a combination of sensors.
cyclists which have such communication capabilities.
On the other hand, point cloud maps are generated by 3-D A. GPS/IMU Based Techniques
scanners, such as LiDAR sensors, and represent the external Global navigation satellite systems (GNSS) such as GPS,
surface of an object in the 3-D space. The data are stored in GLONASS, BeiDou, and Galileo rely on at least four satellites
X, Y, and Z format for each co-ordinate where localization to estimate global position at a relatively low cost. Typical GPS
based on point cloud is performed using techniques such as average accuracy ranges from few meters to above 20 m [20],
Markov localization systems and simultaneous localization and limiting its implementation to autonomous vehicle applications
mapping (SLAM) systems [16]. In the former systems, a vehi- as decimeter level (∼30 cm) accuracy is demanded to stay
cle collects information about the environment and other road in lane [21]. However, GPS accuracy can be improved upon
users from on-board sensors and V2X information and uses the by using differential GPS (DGPS), assisted GPS (AGPS), or
information to localize the vehicle inside a pre-existing static real-time kinematic (RTK) solutions.
feature map while, in the latter ones, LDMs are created as A DGPS utilizes measurements from an on-board vehicle
the vehicle passes the environment and are then compared to GPS unit and from a GPS unit on a fixed infrastructure unit
the pre-existing static map to localize the vehicle. The advan- with a known location as shown in Fig. 3. As GPS error is
tage of SLAM is the utilization of an LDM which means that correlated between two nearby GPS units, calculating this cor-
maps can be updated with each pass of the environment and related error can be used to eliminate the error of on-board
new maps can be created when the vehicle is in a nonmapped vehicle GPS to improve its accuracy. DPGS uses the known
environment. Fig. 2 shows an example of cloud point map that position of the fixed infrastructure unit to calculate the local
can be obtained from a LiDAR, on-board of a vehicle. error in the GPS position measurement periodically, as illus-
trated in Fig. 3. This correction is then broadcasted to the
III. S ENSOR -BASED L OCALIZATION T ECHNIQUES on-board vehicle GPS units to adjust their own GPS estimate
These techniques rely only on on-board vehicle sensors to achieve an average accuracy in the range of 1–2 m [19].
to find the global position of a vehicle in a specified Also, AGPS offers increased accuracy and signal availability
at an increased cost. AGPS uses information from a cellu- technique was successful in mitigating some of the weaknesses
lar network to reduce delays in obtaining a position from the of the standalone GPS and IMU methods, the magnitude of the
satellite as well as increase the signal coverage by reducing localization errors means that the system would be inadequate
acquisition thresholds [22]. However, it provides less position for autonomous vehicle systems. However, the results show the
accuracy compared to a DGPS unit. On the other hand, an potential of fusing data from multiple sensors to improve the
RTK-GPS utilizes dual-frequency GPS receivers to estimate accuracy and robustness beyond what each sensor can achieve
the relative position of a vehicle with respect to the position as standalone.
of a base station with known position. In this case, the rela-
tive position is estimated based on carrier phase measurement
of GPS signals to achieve centimeter level accuracy. Further B. Camera-Based Techniques
improvements in accuracy could be achieved in the future with As a method of replacing GPS with an alternative on-board
the use of the European Galileo GNSS, which is expected to sensor, Li et al. [27] proposed a low-cost localization method
be fully operational with a constellation of 30 satellites by utilizing only cameras, where the images obtained from the
2020 [23]. The combined GPS-Galileo system is intended to cameras were down-sampled to a resolution of 800 × 600 pix-
provide centimeter-level accuracy for GNSS solutions with els to reduce computation time. This vision-based localization
dual-receiver systems [24]. Also, the increased number of approach combines a topological map and a point-cloud map
available satellites in combined GPS-Galileo systems could to provide an SLAM type technique. The localization, first,
improve signal availability. estimates a rough position through dividing the images into
Although DGPS, AGPS, and RTK-GPS methods improve grids and extracting the orientation histograms of each cell.
the accuracy of a conventional GPS, they suffer from poor reli- Then, a fine localization was done using the map consist-
ability. Therefore, the signal availability remains a problem, ing of landmarks in the environment. The proposed two-stage
especially in urban environments, thereby reducing imple- localization method not only increases the accuracy of the
mentation potentials of GPS technologies for the autonomous localization but also reduces the computation requirements for
vehicle applications. Additionally, the time-to-first-fix (TTFF), the fine localization. Using the proposed method, mean posi-
which refers to the time required to obtain the signal from the tioning errors of 75 cm were achieved. However, the system
satellites and acquire an initial position estimate after a GPS is sensitive to changes in illumination conditions or angle of
unit is turned presents another limitation. Typically, TTFF observation which may cause the system to fail. Possible solu-
for a GPS unit can be up to 12.5 min if the unit was com- tions to these weaknesses which were suggested by the authors
pletely turned off. However, this can be partially mitigated by: are the use of omnidirectional cameras with a 360◦ field of
1) keeping the clock operational whilst the unit is off; 2) uti- view to mitigate the field of view change errors and designing
lizing a stand-by mode; or 3) acquiring the satellite almanac more robust algorithm against severe weather and illumination
information from a cellular network with AGPS methods [25]. conditions. An alternative approach of using cameras is visual
For the aforementioned reasons, GPS needs to be either odometry discussed in [28], which uses information from cam-
integrated with other sensors or replaced with more accu- eras to calculate the trajectory of the vehicle for relative posi-
rate sensors in the context of autonomous vehicles. For this tioning. The method suggested by the authors uses stereovision
purpose, GPS is often integrated with IMUs to design a low- provided by two cameras mounted on the front of a vehicle.
cost vehicle localization system. IMUs use a combination of The attitude of a vehicle is calculated incrementally by com-
accelerometers and gyroscopes to measure linear accelerations paring relative movement and orientation to tracked feature
and vehicle angular velocities, respectively. This information points between left and right cameras at different time frames.
can be used to calculate the trajectory of the vehicle as it trav- The algorithm used by the authors is a random sample consen-
els. For this reason, IMUs have been used to estimate vehicle sus algorithm, which rejects outlier estimates to provide robust
position relative to its initial position, in a method known as estimates. Two experiments were completed in an urban envi-
dead reckoning. However, the main problem with IMUs is ronment with cameras recording 30 frames/s at a resolution
accumulated errors where the measured position drifts further of 320 × 240 pixels. Using this method, estimated and actual
away from its true position as the vehicle travels. This problem travel lengths of 145.37 m and 165.86 m in the first experi-
can be overcome by correcting the estimated position using ment and 197.89 m and 216.3 mm in the second experiment
other sensors, to avoid accumulated drift and to provide global were observed, respectively. Main source of errors occurred
positioning. To address the low accuracy and interference when the subject vehicle was stationary while other objects
issues in GPS as well as the cumulative errors related to IMU moved in the vicinity, therefore potential improvements would
sensors, Zhang et al. [26] proposed integrating the informa- be the ability to differentiate between moving and stationary
tion from the GPS and IMU. To test the approach, the authors objects.
used a GPS/IMU system which provides GPS data, heading To further improve the performance of the camera-based
angle, and velocity of the vehicle at 10 Hz. The proposed localization techniques, the integration of GPS, IMU, and
method was successful in increasing the accuracy beyond stan- camera sensors have been proposed by the researchers. For
dalone GPS or IMU capabilities; however, cumulative errors instance, Kamijo et al. [29] suggested using GPS/IMU for
were still present in the system. Over a driving distance of global positioning, whilst using the camera to recognize lane
408 m, the system accumulated root mean square (RMS) markers for lateral positioning. Using this approach, mean
errors of 7.2 m, compared to that of 22.3 m and 13.2 m of positioning errors of 0.73 m were achieved. Suhr et al. [30]
IMU odometry and GPS, respectively. Therefore, while this extended the method proposed [29] by the use of cameras
for recognition of road markers as well as lane markers to traditional pulse-based Radar systems. For example, the K2pi
support both lateral and longitudinal positioning. Over two microwave Radars based on FMCW can offer a 360◦ field of
experiments, this approach obtained mean lateral errors of view and a range up to 100 m, with power requirements as
0.49 m and 0.58 m and mean longitudinal errors of 0.95 m and low as 0.1 W, however, the accuracy is typically lower than
1.43 m for the first and second experiments, respectively. The that of pulse-based radar systems [32].
authors noted that the method has larger longitudinal errors As an example of localization system using a Radar,
than lateral errors because lanes and road markers were used Vivet et al. [32] evaluated the data obtained from a 360◦ field
for lateral localization, whereas longitudinal localization only of view FCMW microwave Radar sensor through two differ-
uses road markers. Since lanes are observed with a higher ent SLAM methods to localize a vehicle. The first method
frequency than road markers, the proposed method experiences is a trajectory-oriented SLAM technique while the other one
higher longitudinal errors compared to lateral errors. Major analyses the distortion caused by rotating Radar at high speed
errors were observed: 1) when driving on long straight roads to obtain the trajectory of the vehicle and map the envi-
with no road markers detected; 2) when the vehicle was pass- ronment. For a vehicle traveling at the speed of 30 km/h,
ing through intersections with no road markings; or 3) when the methods resulted in mean position errors of 10 m and
false detections of road markers occurred. As another example 12 m for the first and second techniques, respectively, thereby
of integrating vision-based techniques with GPS and IMUs, indicating the technique will be unsuitable for autonomous
Mattern and Wanielik [16] proposed two different approaches vehicles. However, the work in [31] explored the use of pulse-
for localizing vehicles based on combination of on-board vehi- based SRR due to its low cost and good accuracy, where the
cle camera and aerial images along with IMU/GPS signals. Radar sensor acquires up to 64 detections, each at 20 Hz.
The first method utilizes feature maps based on aerial imaging Also, information of speed and yaw rate were used from sig-
containing information about landmarks, lane markings, curbs, nals of a GPS/IMU system. The results showed RMS errors
and the road geometry. Imaging from the camera equipped on of 7.3 cm laterally and 37.7 cm longitudinally, with worst
the vehicle was then used to localize the vehicle within the case errors of 27.8 cm laterally and 115.1 cm longitudinally.
feature map. The second approach uses aerial images which While these results are promising for Radar-based localiza-
are then processed to remove unnecessary information so that tion, Radar maps are dependent on the quality of discernible
information only about local edges (e.g., the edges of features features available, which can cause errors when such fea-
such as roads) is retained. The experimental results showed tures are not available. To further improve the accuracies of
that, for the first method, 80% of position estimates had both Radar-based techniques, a novel approach of utilizing ground
lateral and longitudinal errors less than 1 m, while the sec- penetrating Radar technology for localization was proposed
ond method only achieved 60% of position estimates within in [13]. This method scans the subsurface features and the
this range. Moreover, the second method had peak errors of inhomogeneity of the subterranean geology to create a map.
3.5 m and 7 m which would cause the vehicle to choose the These features are unique and static enough that localiza-
wrong lane. The authors mentioned that future improvements tion for autonomous vehicles could be completed utilizing
to these techniques could be achieved by further processing subterranean feature maps. Testing was done using a vehicle
of the aerial or camera imaging. Overall, these two methods equipped with GPS/IMU system integrated with the localizing
are still inadequate for autonomous vehicle localization due to ground penetrating Radar (LGPR) system with a ground pen-
the magnitude of errors present. etrating depth of 2–3 m, which was brought up to speeds of
In general, although cameras can offer low cost solutions 100 km/h in testing. The vehicles first created a subterranean
to localization systems, the level of the accuracy depends on feature map of the environment over the initial pass of the envi-
the image resolutions. It is noted that acquiring high resolution ronment and then attempted to localize itself within this map.
images can drastically increases the processing power demand Results showed the capability to localize within positional
and processing time of the localization system which cannot RMS errors of 4 cm, which is within limits for a vehicle to
be achievable using on-board vehicle processing systems. maintain its lane of traffic. The proposed approach, therefore,
shows significant potential as it can provide the high accu-
C. RADAR-Based Techniques racy localization required for autonomous driving. Moreover,
unlike localization systems such as camera-based localization
A Radar sensor is a ranging sensor which utilizes radio
systems, LGPR is not affected by weather or lighting con-
waves. Radar functions by emitting periodic radio waves
ditions, thereby making it more robust to different operating
which can bounce off obstacles back to the receiver and dis-
conditions. However, the authors pointed out that this concept
tance to target is measured from the time of arrival of radio
is relatively new and still requires further study to understand
waves. Each radio wave provides a single range measurement
its capabilities and limitations.
which gives the distance to the obstacle that reflected it back to
the receiver. Radars also have relatively low power consump-
tion, for example, the Delphi short range radars (SRRs) use D. LiDAR-Based Techniques
only 0.16 W and offer up to 64 range measurements at 20 Hz It is noted that Radar maps are susceptible to errors in the
with a field of view of ±75◦ and a range of 80 m [31]. Even case of changes in the pre-existing map due to their lim-
lower power requirements can be achieved by frequency mod- ited capabilities in collecting environmental data. Therefore,
ulation continuous wave (FMCW)-based Radars, which use to increase localization accuracy and robustness, more accu-
continuous Radar signals rather than the periodic ones used in rate maps with denser point clouds are required. LiDAR
on static maps, which meant that extreme changes to the

road environments could cause the technique to fail. Possible
solutions to this weakness were suggested such as using prob-
abilistic maps to identify changed environments, comparing
GPS and map-based localization to spot localization errors or
creating 3-D environment models beyond the road surface.
Therefore, to improve robustness to dynamic environments,
the technique was further extended in [36] by incorporating
the use of LDMs. Instead of creating orthographic maps of
Fig. 4. Principle of LiDAR sensor. The sensor sends a laser signal and upon the road surface with fixed infrared intensity values, a method
receiving the signal back obtains the infrared reflectivity of the object as well
as distance to the object based on time for the signal to bounce back.
of using probabilistic infrared intensity values collected over
multiple passes of the same environment was used. The use of
probabilistic maps increases robustness to changes in the envi-
technology can collect significantly more data than Radar sen- ronment and allows the system to identify dynamic objects.
sors, and therefore potentially offer higher accuracy compared In experimental testing, a GPS/IMU system updating inertial
to the Radar-based techniques. A LiDAR sensor measures information and global position at 200 Hz together with a
distance to a target using multiple laser beams which each 64-beam LiDAR sensor providing a million 3-D points per
measures the distance to the target, based on the time of arrival second were used. A large amount of acquired data enabled
of the signal back at the receiver, as well as the infrared inten- the system to scan map cells multiple times and create
sity of the obstacle, as illustrated in Fig. 4. LiDAR systems a probabilistic map using computation of intensity values. The
can be used to collect large amounts of ranging and infrared proposed method was shown to achieve positional RMS error
data from obstacles, neighbor road users, etc. For instance, of 9 cm, suggesting accuracy feasible for urban autonomous
the Velodyne-64 system, which is a common LiDAR sensor driving as well as improved robustness to dynamic environ-
used for localization systems, collects approximately 2.2 mil- ments compared to the previous methods.
lion 3-D points and their associated infrared intensity values Further improvements to the accuracy of LiDAR-based
per second over a range of 120 m and a field of view of 360◦ localization was presented in [37]. The authors utilized an
horizontal and ∼ 27◦ vertical [34]. However, LiDAR technol- alternative grid representation of the ground which is invari-
ogy is very expensive compared to other sensing technologies ant to the laser-perspective or vehicle motion. This eliminates
and its performance is sensitive to environmental conditions the need for postfactory reflectivity calibrations required by
such as rain or snow. The high-power requirement is also other methods, improving localization performance. Using
a major disadvantage of LiDAR sensors. For example, the the proposed technique, longitudinal and lateral localization
Velodyne-64 system uses as much as 60 W to power itself. RMS errors of 3.3 cm and 1.7 cm were achieved, thereby
To investigate the use of LiDAR for localization, the work improving over the performance of previous state-of-the-art
done in [35] suggested using LiDAR to detect curbs and road LiDAR localization methods without increasing the computa-
markings to create a feature map of the environment and local- tional requirements. Besides increased accuracy, an additional
ize vehicles within the map. In the proposed approach, curbs advantage of using LiDAR is the increased robustness to
were identified by acquiring LiDAR measurements in 32 con- changes in the mapped environment when compared to tech-
centric rings and analyzing the distance between the rings to niques such as Radar or camera-based localization techniques.
identify curb-like features using filters. Also, road markings For instance, in [38], a vehicle equipped with LiDAR, IMU,
were identified by analyzing the LiDAR reflective intensity and wheel odometry successfully drove autonomously in an
data and comparing it to expected values for road markings. urban environment. The proposed system localized itself while
These two features were then used to localize the vehi- driving autonomously on a 1.9 km long route utilizing point
cle within the feature map. For the urban experiments, an cloud data collected six months ago and data fusion via a par-
RTK-GPS, an IMU and a LiDAR sensor with 32 beams ticle filter algorithm. The resulting mean longitudinal and
and maximum detection distance of 33 m were used. The lateral localization errors were 8.2 cm and 7.5 cm, respec-
proposed approach resulted in lateral and longitudinal errors tively. Therefore, highly accurate localization was still possible
of less than 30 cm which were considered satisfactory for even when changes had occurred in the environment and
autonomous driving in urban environments. In comparison, localization was performed in a different season from the
Levinson et al. [21] proposed a solution integrating GPS, initial mapping data.
IMU, wheel odometry, and LiDAR to generate high-resolution While LiDAR techniques can offer very accurate and robust
maps. The authors suggested eliminating map features that are localization methods, the drawback is their high power and
unlikely to be static to create a 2-D map of the road sur- computational requirements as well the high implementa-
face in the infrared spectrum. Therefore, obstacles such as tion costs. This presents a problem for mass production and
moving cars are eliminated from the map. A SLAM-style marketing of autonomous vehicles equipped with LiDAR sen-
relaxation filter was used to localize the vehicle within the sors. Attempting to reduce the computational requirements
created map at 200 Hz. Using the proposed approach, errors of LiDAR-based localization, large point cloud maps were
as low as 10 cm were obtained, although, in some occasions, reduced to compact Gaussian mixture maps in [39]. While
such as when turning, errors reached as high as 30 cm. The raw point cloud maps can require up to 500 MB of data per
main weakness pointed out in the technique was its reliance km of road, the Gaussian mixture maps require 10–100 MB
of data, depending on the number of Gaussians per grid cell. potentials and limitations. Suitability of the techniques for
Using the Gaussian mixture maps for localization resulted in autonomous vehicles is based on the robustness and relia-
longitudinal and lateral RMS errors below 10 cm and 13 cm, bility as well as capability for in-lane localization accuracy.
respectively. Moreover, the proposed method was shown to The required accuracy for in-lane localization is taken as
be robust to harsh weather conditions and changes in road 30 cm, as derived in [4]. In summary, the analysis shows
appearance due to construction work. On the other hand, that no sensor as standalone offers adequate accuracy and
addressing the high implementation cost of LiDAR techniques, robustness required for autonomous driving and data fusion
Wolcott and Eustice [40] investigated the use of camera-based of multiple sensors has significant potential to design a cost
localization within pre-existing LiDAR maps. In contrast to effective localization system meeting accuracy requirements
LiDAR, camera-based technology is less accurate and is sus- for autonomous driving. Integrated GPS/IMU/camera localiza-
ceptible to changes in illumination conditions or angle of tion systems provide accuracy up to 73 cm, however, further
observation but is significantly cheaper. Therefore, the authors improvements are needed to offer the accuracy and robustness
suggested creating initial maps used for localization using required for fully autonomous vehicles. Radar sensor-based
LiDAR sensors and equipping autonomous vehicles with cam- techniques offer cheaper localization systems compared to
eras to localize themselves within the LiDAR maps. This LiDAR-based systems and can meet the accuracy require-
means that the highly accurate LiDAR maps are utilized, but ments for autonomous vehicles, as shown in [33]. However,
autonomous vehicles could be significantly cheaper. This tech- the robustness of these methods remains an obstacle for imple-
nique was shown to localize with longitudinal and lateral RMS mentation. In contrast, LiDAR can offer high accuracy but
errors of 19.1 cm and 14.3 cm, respectively, with data captured at a significantly higher cost compared to Radar. Therefore,
at 10 Hz, which provides a similar order of magnitude errors to for LiDAR to be a commercially feasible option further tech-
LiDAR techniques but at a significantly reduced cost, power, nological advances would be required to reduce the cost or
and processing requirements. Alternative methods of utilizing alternatively the approach used in [40] could be used to take
laser technology whilst maintaining low implementation costs, advantage of the high accuracy and robustness of LiDAR-
is the use of single beam laser range finders (LRFs), such as based maps, but keep the cost of autonomous vehicles low by
in [41], where A GPS system, gyroscope, two LRF systems equipping them with cameras instead of LiDAR sensors. This
and a 2-D feature map, consisting of road and building shapes, type of approach could be the key to achieving high accuracy
were integrated. The two LRFs scanned the environment, with and formulating low-cost solutions but robust performance in
one scanning horizontally and one vertically to identify build- different environment conditions is still a challenge due to
ing facades and build a feature map based on this information. limitations of camera systems.
Comparing the pre-existing feature map to the LDM resulted
in mean positional errors of 3.098 m, which is unsuitable for IV. C OOPERATIVE L OCALIZATION T ECHNIQUES
autonomous vehicle localization.
The augmentation of off-board information obtained
through V2V and V2I communication systems to the sensory
E. Ultrasonic-Based Techniques
information has shown the potential to improve the vehicle
Other localization methods have attempted to use alterna- localization accuracy, robustness, and reliability in different
tive low cost sensors such as ultrasonic sensors. For instance, driving and environmental conditions [43]. In such systems,
Jung et al. [42] proposed the use of ultrasonic sensors inte- vehicles can broadcast information about their states to other
grated with a set of sensors including a digital magnetic vehicles (V2V), including speed, heading, and location, as
compass, a gyroscope, and two encoders for ultrasonic-based well as the information related to the environment while
SLAM techniques. Ultrasonic sensors can scan the environ- adverse weather conditions or obstacles can be acquired from
ment by utilizing a mechanical wave of oscillating pressure infrastructure (V2I).
which can propagate through air or other materials. Distance to The cooperative localization techniques use wireless com-
target can be measured based on the time of arrival of the sig- munication devices, such as Wi-Fi, cellular, and UWB radio
nals back to the receivers. Ultrasonic sensors were chosen due communications where transmitted signals are used to esti-
to their high performance with low electric power consumption mate the range to the broadcaster. There are several approaches
and low cost. However, due to inaccurate extracting of fea- to estimate the distance or relative position to the broad-
ture points, the localization process could take very long caster of a signal. To calculate the distance to the broad-
times. Average processing times of 10.65 s were observed, caster only, one radio signal is required, while calculating
thereby making the technique unsuitable for high-speed vehi- a relative position requires three radio signals in case of a 2-D
cle applications. Also, the associated long average processing localization or four in the case of 3-D localization. In gen-
time causes accumulated errors due to measurements from the eral, the positioning and ranging systems can work based on
sensors such as IMU. Moreover, the detection range of the four principles, namely time-of-arrival (TOA), time-difference-
ultrasonic sensors is limited to 3 m which is not sufficient for of-arrival (TDOA), angle-of-arrival (AOA), and radio-signal-
obstacle detection system of an autonomous system. strength (RSS). Fig. 5 illustrates the concepts of TOA, TDOA,
and AOA. In TOA approach, a vehicle initiates the trans-
F. Discussion mission by sending a signal to a reference node, which then
Table I compares the aforementioned localization techniques sends a signal back, and relative distance to the reference node
in terms of sensor configuration, accuracy, and the associated is measured based on time for the response signal to arrive
TABLE I
S UMMARY OF O N -B OARD S ENSOR -BASED L OCALIZATION T ECHNIQUES
back at the vehicle. The accurate ranging of TOA approach signal strength while traveling from node to node is measured
is limited due to the need of a perfect synchronization of to estimate the travel distance of the signal. It is noted that
clocks between all nodes which can be difficult to achieve a radio signal is transmitted from a broadcaster (i.e., a node)
in practice. In TDOA approach, the position of a vehicle which can be stationary (e.g., infrastructure) or mobile (e.g.,
is estimated using transmission signals of three broadcast- vehicles).
ing stations, where one of the stations, known as the master An ultrawide band (UWB) sensor uses radio signals to
station, initiates the transmission by sending a signal to the spread information over a wide portion of the frequency
vehicle and the other stations, shown in Fig. 5(b) as sta- spectrum (>500 MHz), allowing large amounts of rang-
tions B and C. Once stations B and C receive the initial ing data to be transmitted with a low power requirement
signal from the master station, they send a signal to the vehi- for short distances and without interference with other sig-
cle. Since the location of all stations relative to each other nals in the same spectrum [45]. Moreover, a UWB sensor
is known, the travel time of radio signals between them is is capable of penetrating through obstacles due to using
known as well. Based on the time difference of signal arrivals the large wavelengths of the radio signals. A UWB sensor
T AV , T BV , and T CV , the distance to each station and thereby estimates the vehicle relative position to a stationary UWB
the relative position of the vehicle can be calculated [44]. In station through communications between different UWB sta-
AOA approach, as illustrated in Fig. 5(c), antenna arrays mea- tions. However, the limited range of the signals (5–50 m)
sure the angle of the signal at each node. However, the need for requires well-developed infrastructure thereby limiting the uti-
antenna arrays at each node increases the cost and complexity lization of UWB sensors for vehicle localization systems.
of the system. Finally, in the RSS method, the attenuation of Therefore, the sensor is typically used in short range, high
A. Vehicle-to-Vehicle Localization Techniques

In this paper, a V2V localization technique refers to a local-
ization technique which integrates information from adjacent
vehicles with the on-board sensory information. V2V commu-
nication systems form a vehicular ad-hoc network (VANET) of
interconnected mobile nodes which are aware of the locations
and trajectories of each other. Although distance measure-
ments to a VANET broadcaster cannot be as accurate as,
for example, LiDAR measurements, it offers distance mea-
(a) (b)
surements out of line-of-sight and at relatively low cost
and less processing requirements. V2V systems are usually
based on the IEEE 802.11p protocol, which is an extension
to the IEEE 802.11 protocol that defines the standards for
Wi-Fi technology. The IEEE 802.11p-based wireless access
in the vehicular environment (WAVE) technology can offer
bit rates of 3–27 Mb/s over a communication range of up
to 1000 m [48]. The message types expected to be used
in VANET systems can vary regionally, for example, in
(c) the European context, V2V communication systems mainly
depend on the use of three different types of messages:
Fig. 5. Direction-based ranging techniques. (a) TOA method. 1) cooperative awareness messages (CAMs); 2) decentralized
(b) TDOA method. (c) AOA method.
environmental notification messages (DENMs); and 3) ser-
vice announcement messages (SAMs). CAMs are periodically
transmitted to share vehicle information (position, dynam-
data rate applications such as indoor industrial autonomous ics, attributes etc.) to other road users thereby maintaining
vehicles [46]. cooperative awareness in the VANET [49]. DENMs are event
Radio waves can also be used for data communications driven messages containing information about an event, such
in wireless networks, such as Wi-Fi or cellular networks. as an accident or a hard-braking vehicle, to other nearby
Wi-Fi devices are defined by the IEEE 802.11 standards, vehicles [50]. SAMs are messages sent by vehicles or infras-
which specify the set of media access control and physical tructure on the control channel to announce services available
layer specifications that define the Wi-Fi technology. Wi-Fi on the so-called service channels [51].
devices operate in various frequency bands, with a 75 MHz of The widely used technique of V2V systems to improve
spectrum in the 5.9 GHz band reserved for V2V and V2I com- GPS accuracy is multilateration, illustrated in Fig. 6, in which
munications by the European Commission [47]. These devices multiple GPS measurements from adjacent vehicles as well
can be used in vehicle localization systems to receive informa- as distance to the other vehicles are used to increase the
tion from infrastructure or other vehicles as well as transmit overall accuracy of a vehicle’s self-localization estimate. The
information to other vehicles. The networks can either be more position estimates the vehicles can acquire, the more
infrastructure-based with access points and routers or they can accurately they can estimate their own position [52]. In [53],
be ad hoc networks in which each node participates in rout- simulation of vehicles equipped with multilateration (V2V and
ing by forwarding data to other nodes. Wi-Fi devices achieve GPS) systems was carried out to show that V2V communi-
data transmission rates ranging from 6 Mb/s up to 54 Mb/s, cation can reduce average localization errors from 6.75 m for
with communication ranges up to 100 m and transmission GPS-only localization, down to 3.30 m. However, factors such
power requirements of up to 100 mW [48]. On the other as uncertainty of intervehicle distance, the number of vehicles
hand, cellular networks use technologies such as 3GPP long in a cluster, communication latency, and communication fail-
term evolution, i.e., 4G, to transfer data between the cellular ures will affect this performance. During communication
devices. Cellular networks utilize cellular base stations with failures or with high errors in the intervehicle distance mea-
a transmission range of up to 15 km to enable data trans- surements, these errors tend toward the GPS-only localization
mission. Each base station provides communication to a cell error. Similarly, in [52], the localization accuracy of a mul-
around it. As the cellular device travels toward the neighbor tilateration system was improved by integrating information
cell the signal strength from the next base station becomes from on-board sensors, such as millimeter wave Radar. Using
stronger and therefore the device switches communications simulation results and assuming a communication interval of
to that base station in a procedure called hand-off. Although 100 ms and a maximum sensing range of 100 m, the authors
cellular devices typically only use one base station for com- showed that their approach, when all vehicles were equipped
munication, multiple cellular base stations are typically within with GPS and V2V communication devices, achieved an aver-
communication range, which enables localization through the age error of 0.6 m. Therefore, the proposed approach reduced
ranging methods discussed previously. Cellular networks typ- the localization error by 85% compared to GPS solutions and
ically broadcast in the 800, 900, 1800, 2100, or 2600 MHz 60% compared to a multilateration system. In the research,
bands, achieving transmission rates up to 100 Mb/s [48]. while the accuracy intervehicle distance estimation depends
AOA measurements to improve localization accuracy. Also,

the proposed approach used vehicle state information from
a gas pedal, brake pedal, and steering wheel sensors. In this
method, the vehicles used the shared state information from
other vehicles to gain a better estimation of their own location
as well as estimate the location of the other vehicle and then
Fig. 6. V2V multilateration. Transmitted positions and calculated distances
to the nearby vehicles are used to obtain additional self-localization estimates
sent this information back to them. Therefore, vehicles not
for subject vehicle A. only gain information regarding other vehicles in the network,
they acquire more estimates about their own position as well.
The proposed approach was tested in simulations with five
on the number of vehicles with V2V communication systems, vehicles and achieved a mean square error (MSE) of 0.52 m
it was shown that if 100% of the vehicles are equipped with for the best performing vehicle and 1.65 m for the worst one.
a V2V technology, the system recognized 70% of all nearby Another approach for improving multilateration accuracy is
vehicles within an error of 1 m while if 40% of vehicles were weighted average localization. This approach assigns confi-
equipped with V2V systems, only 50% of all nearby vehi- dence values to each position estimate received, and gives
cles within an error of 1 m were recognized by the system. higher weight to measurements with higher confidence in the
Therefore, the relationship between the number and penetra- localization algorithm. For example, Ahammed et al. [57]
tion rate of connected vehicles and localization accuracy is proposed a method with weights determined based on distance
nonlinear where a minimum number of connected vehicles to the broadcaster, since vehicles further away from the subject
would be required for adequate accuracy and reliability. Also, vehicle are more likely to produce erroneous distance measure-
the results of the simulations show that the V2V localization ments. The proposed method was evaluated with a network
paired with on-board ranging sensors can achieve good accu- of 10 cars, resulting in a mean localization error of 2.38 m.
racy. This approach was also evaluated in field tests, in [54], This approach was further extended in [58] to include signal
where six vehicles with GPS/IMU and V2V systems were to noise ratio (SNR) in the determination of weight factors as
tested with one vehicle with a highly accurate positioning well, since signal noise can affect the localization accuracy.
system to be used as the ground truth. Measurements from The approach was evaluated through multiple simulations with
two of the six vehicles showed unexpected behavior and were network sizes ranging from 20 to 200 cars. The localization
therefore removed from the results. The remaining results errors observed ranged from 85 cm for 20 cars down to 25 cm
showed that 50% of the errors were within 1.09 m, result- for 200 cars. This shows that the inclusion of confidence values
ing in an improvement of 66% compared to GPS as well as in a multilateration system can greatly improve accuracy.
95% errors were within 3.54 m, resulting in an improvement It is important to note that all abovementioned cooperative
of 85% compared to GPS. Therefore, the accuracy of VANET localization techniques rely on the use of GPS for the vehicle
localization is highly dependent on the number of connected self-localization. Whilst GPS may give adequate performance
vehicles in the area, as well as the accuracy of the shared posi- in most scenarios, as mentioned before, GPS suffers from poor
tion estimates. Moreover, the unexpected behavior of the two reliability in some environments such as dense urban envi-
vehicles in the experiment highlighted the need to mitigate ronments. Therefore, in safety critical applications such as
the possible effect of erroneous inputs from the other vehi- autonomous vehicles, the localization system would need to
cles. Therefore, the localization algorithm would need fault be fault tolerant in order to continue operating with adequate
detection and isolation to reject any erroneous inputs. performance even in the case of high GPS errors or signal
Ordóñez-Hurtado et al. [55] extended multilateration meth- blockage. Simply integrating the GPS with other on-board sen-
ods to include the use of stationary vehicles for improving the sors would be a possible solution to offer reliability through
localization accuracy. Since stationary vehicles can localize sensor redundancy [59]. Other considerations for the reliability
themselves more accurately given enough time, the authors of V2V localization systems are the network quality of ser-
propose using stationary vehicles (e.g., parked vehicles) to vice. In [47], it was noted that the limits of the 802.11p can act
transmit their location via V2V, so nearby vehicles can gain as a barrier to situational awareness since the 5.9-GHz band
a more accurate self-localization estimate. Although battery used is very sensitive to obstacles. While in good communica-
discharge of a powered-off vehicle is a concern, the authors tion condition the authors could establish a link from 700 m,
pointed out that using a 1-W on-board unit, a 480 Wh bat- a single thin wall of trees could create as much as 25% packet
tery can last up to 2 days. The method was evaluated using loss with distances as small as 50 m. Therefore, the effect of
a small-scale and a large-scale scenario with 20 and 900 cars, highly congested networks with fast moving nodes and vary-
respectively. Compared to multilateration without stationary ing topology on the performance on cooperative localization
vehicles, the simulation results showed up to 55.09% and systems will need to be considered.
62.85% improvement in the small-scale and large-scale cases, Table II shows a summary of the V2V localization meth-
under best scenarios with the highest number of parked cars. ods discussed above. Currently, the most researched V2V
The average improvement for both test cases was 33.01%. techniques are multilateration systems which utilize GPS and
This shows the potential improvement in accuracy from the V2V communication. The results in Table II show that these
use of stationary vehicles in V2V systems. Furthermore, techniques can greatly improve GPS localization accuracy.
Golestan et al. [56] used multiple GPS position estimates However, while the accuracy of V2V localization systems
from other vehicles and ranging estimates using TOA and based on multilateration are not adequate for autonomous
TABLE II
S UMMARY OF V2V L OCALIZATION M ETHODS
vehicles, they show great promise for integration to more is the large number of RSUs required for continuous commu-
accurate on-board sensor-based techniques (e.g., camera-based nication, which will require large initial installation works and
or LiDAR-based) to further improve performance, accuracy, costs. Therefore, to limit these initial costs the design of V2I
reliability, and robustness at a relatively low cost. The main communication systems should be efficient and reliable with-
limitation of V2V localization techniques is their reliance on out needing excessive infrastructure implementation. As such,
high penetration rate and number of connected vehicles which the spacing and configuration of RSUs must be studied, as
cannot always be guaranteed to be available. well as different communication methods and their effect on
system parameters (e.g., networks latency, packet loss, etc.).
As shown in [60], output power transmission, antenna height,
B. Vehicle-to-Infrastructure Localization Techniques and tilt, the length of packets and the access mechanism,
In vehicle-to-infrastructure (V2I) systems, the vehicles com- all affect the performance of V2I communication networks.
municate with RSUs or cellular base stations at fixed positions. Also, optimal configurations of roadside beacons may vary
This allows the RSUs to communicate their location, weather under different traffic loads. Further, the use of existing
conditions, traffic flow, etc. with vehicles to estimate their own cellular infrastructure can be used in order to reduce capital
position more accurately. The potential advantages of the V2I expenditure.
systems over V2V systems include: 1) offering a more accu- There are multiple methods for estimating the relative posi-
rate location estimate due to fixed position of RSUs (similar to tion from the broadcaster of a signal as discussed previously.
the stationary cars in the V2V localization, but more reliable For instance, Hassan et al. [61] proposed the use of impulse
as the RSUs are fixed rather than only temporarily stationary); radio UWB (IR-UWB) signals for V2I localization and com-
2) improved reliability as the number of RSUs within trans- pared different methods for relative positioning to the broad-
mission range can be guaranteed unlike in V2V techniques; caster. The main benefits of IR-UWB are low-transmit power,
3) broadcasting critical information (e.g., events such as acci- low cost, wide bandwidth usage, high precision, and resistance
dents which have occurred in the area or adverse weather to multipath errors due to short pulses. The proposed system
conditions), allowing the vehicles to have a greater awareness uses V2I communication devices placed at lampposts at 20 m
of their environment and allowing them to react to any risks; intervals to broadcast their position to nearby vehicles. Based
and 4) enhancing traffic flow and safety through supporting on the distance to the RSUs and the distance measured from
vehicle control and traffic management systems where vehicle the RSU, the vehicle can estimate its position. The authors also
speed, steering, and route selection are controlled by a control discussed various methods such as TOA, TDOA, and AOA for
station aware of other vehicles on the road. However, the draw- estimating the distance between the vehicle and RSUs. In sim-
back of V2I systems based on the IEEE 802.11p technology ulations, TDOA was observed to be the most efficient at the
middle of the network, whilst AOA was the most efficient on a highway with base stations deployed at 200 m intervals.
at the edges of the network. Therefore, the authors suggest The simulation results showed that at 50 and 100 MHz sig-
a hybrid technique utilizing TDOA and AOA techniques would nal bandwidth the position accuracy was below 25 and 20 cm
be optimal for V2I applications. in 99% of the cases, respectively. These results showed that
In contrast, Fascista et al. [62] proposed a localization 5G networks can offer very accurate positioning, adequate for
technique based on AOA estimation. In the proposed system in lane positioning. Moreover, the use of existing infrastruc-
beacon packets from an RSU were used to combine the ture, such as lampposts on a highway, can reduce the cost of
AOA of the signal and the location of the RSU broadcasted deploying the 5G base stations.
in the signal into a least squares algorithm to localize the In general, the main drawback of V2I techniques is the high
vehicle. The system was tested in a simulation and showed cost of implementing new dedicated infrastructure. A possible
very good accuracy when close to the RSU, due to high SNR solution to mitigate this problem is utilizing existing cellular
and favorable geometrical conditions. The performance, how- infrastructure for localization. A cellular device can generally
ever, degraded as the distance between the vehicle and RSU communicate with multiple cellular base stations at any given
increased, with GPS-level accuracy obtained at approximately time, which enables techniques such as RSS to estimate the
155 m away from the RSU. However, with closer spacing of position of the phone based on the distance to the base sta-
the RSUs, this weakness could be mitigated. The benefit of tions. There are two approaches to utilizing RSS measurements
this technique is that no sensors are required on-board the for cellular-based localization: triangulation and fingerprinting.
vehicle apart from V2I communication systems, therefore, Triangulation uses path-loss models to estimate the distance
the main cost of the system would be the implementation to multiple base stations and then the vehicle location can
of the RSUs. On the other hand, Khattab et al. [63] sug- be calculated based on the location of each base station and
gested V2I localization system for a vehicle equipped with the estimated distance to them. However, RSS-based triangu-
IMU and V2I communications. Localization is based on two- lation has been shown to provide limited accuracy due to the
way TOA ranging with an RSU and vehicle kinematics from inaccuracy of the path-loss models [66]. On the other hand,
the IMU, which achieved RMS errors of 1.8 m. An alternative in fingerprinting the signal characteristics from base stations
approach using infrastructure to provide lateral positioning for are recorded at different locations and stored in a database.
in-lane localization was presented in [64], where the authors The experienced signal characteristics can then be compared
used passive transponders integrated into lateral white strips to the reference values in the database to estimate the posi-
on the roads and on-board units to detect the transponders. tion of the cellular device [59]. While fingerprinting methods
These passive transponders, unlike the RSUs in other meth- have been successful in indoor environments with up to 5 m
ods, do not broadcast a signal but only reflect the signal from accuracy [67], average errors up to 94 m were seen in an
the on-board sensors back to the vehicle. The benefit of the outdoor urban testing [68]. Alternatively, fingerprinting could
proposed approach is a very low cost of the transponders be based on signals from Wi-Fi access points, which has
and large penetration depth in water and snow due to the been shown to achieve localization errors of 18 m and veloc-
ultrahigh frequency band utilized. The proposed method was ity estimation errors of 25% [69]. However, in [70], a novel
tested in several laboratory experiments including in presence approach to fingerprinting was presented in which a vehicle
of parasitic reflectors, asphalt, and other transponders. The was equipped with 16 GSM devices to scan different GSM
results showed very high accuracy with errors below 3 cm channels in parallel, GPS for global positioning, as well as
in all experiments. However, the low range of the utilized IMU and magnetic compass for trajectory estimation. The RSS
method is a drawback; all experiments were completed for values of each channel were then recorded on different roads
distances below 2 m, and the results showed increasing error and combined with the trajectory estimation to localize the
with distance. Therefore, the transponders would need to be vehicle based on these values. The authors utilized a two-stage
deployed at small intervals and a large quantity of them would localization technique where a coarse localization was first
be required, although their low cost helps mitigate this draw- done based on GPS data, and then a fine localization based
back. However, further testing would be required in a scenario on the GSM RSS characteristics and vehicle trajectory esti-
with the transponders deployed on the road with all the ade- mation. Average localization errors of 4.2 m with 90% errors
quate protection required and with moving vehicles to validate within 5.3 m were achieved using this approach. However, one
these high accuracies and investigate the robustness of the drawback of the method is the update requirements of the RSS
system to more diverse operating conditions. characteristics, as the average localization error increases to
It is noted that communication systems for advanced driver 4.6 m when using six-week-old data. Therefore, the localiza-
assistance systems or autonomous vehicles need to have tion accuracy could significantly degrade over time if updates
low latency, high accuracy, and strong security. The future are not completed at regular intervals, which would increase
fifth-generation (5G) mobile networks are expected to offer the cost of the system. However, this approach would not
increased reliability, reduced latency, and higher throughput require new infrastructure, therefore the main associated costs
compared to current mobile networks, showing great promise would be equipping the vehicles with the necessary sensors
for an alternative approach to current vehicle communication and updating the GSM scan databases.
systems [4]. Del Peral-Rosado et al. [65] studied the feasibility Techniques integrating V2V, V2I, and on-board sensors
of 5G networks for positioning using 5G-like multicarrier sig- have also been investigated by researchers. For instance,
nals and TDOA measurements. The considered scenario was Bento et al. [71] combined V2V/V2I communication with
GPS, IMU, magnetic sensor, and laser scanner. The proposed perceivable to the human eye [81] supporting data rates up to
method used magnetic markers positioned at intersections 96 Mb/s [82]. This enables transmission of data in V2I appli-
and/or roundabouts, of which the known location was cation via traffic and street lights with ranges of 50–100 m [83]
broadcaster by an infrastructure agent to nearby vehicles. and in V2V applications via headlamp and taillights with
Additionally, GPS/IMU was used for global positioning, whilst ranges of 40–50 m [80]. VLC has been shown to provide
the laser scanner was used to provide range and direction good localization accuracy in indoor environments, such as
estimates of the vehicles in front. This technique achieved Epsilon [84], which utilized RSS and trilateration to provide
RMS errors of 1.03 m, which still has room for improvement localization with 90th percentile accuracy of 0.4 m in an
but shows the potential for the combination of V2V and V2I office environment. However, adapting VLC to outdoor vehic-
communications systems for providing low-cost and accurate ular applications introduces further complications, such as
vehicle self-localization. Similarly, in [72], V2V/V2I commu- signal attenuation from severe weather conditions and ambi-
nications were integrated with GPS. In addition to localization, ent noise from direct sunlight or other strong light sources.
the authors attempted to provide improved yaw, pitch, and roll Robustness to ambient noise can be improved by utilizing
estimates over traditional GPS/IMU systems. In the proposed higher frequencies and high pass filters. Yu et al. [80] inves-
architecture, there are three main components: 1) road vehi- tigated the localization accuracy of VLC in a V2V setting at
cles; 2) RSUs; and 3) master antenna fixed stations (MAFS). multiple signal frequencies through numerical analysis, claim-
The road vehicles are equipped with two GPS receivers posi- ing that accuracies below 1 m and 0.1 m could be achieved by
tioned longitudinally on the vehicle’s roof. The RSU was headlamps with signal frequencies of 20 MHz and 40 MHz,
equipped with a GPS receiver and V2I communication system. respectively. It should be noted that increasing the signal
The MAFS was equipped with a GPS receiver in a fixed frequency has the drawback of increasing processing require-
position with good signal connectivity. The MAFS then broad- ments and degrading transmission power. Also, simulation
casts its position to the other units. The system was evaluated results by Kim and Jung [85] suggested that submeter local-
using experiments which showed that the system is capable ization accuracy could be achieved when within 20 m of LEDs
of self-localization within RMS errors of 0.698 m and stan- in a tunnel environment where both tunnel ceiling lights and
dard deviation of 0.14 m. Moreover, the system showed good vehicle taillights were utilized for trilateration. However, the
performance in yaw and pitch estimations, obtaining mean main drawback of VLC when compared to DSRC is the direct
errors of 7.90◦ and 0.86◦ , respectively. However, roll estimates line-of-sight requirement and lower communication range. In
showed large errors, with the mean error of 40.70◦ . The high an urban scenario simulation, Yu et al. [80] showed that
roll estimate error is due to roll estimates requiring collinear DSRC could communicate with an average of 32 vehicles,
vehicles that are close to each other, however, when the vehi- while VLC performed significantly worse by communicating
cles are nearly collinear the estimates are very sensitive to only with an average of five vehicles in the same scenario.
noise, leading to high errors. Nevertheless, VLC could be well suited to some environments
Similarly, combining V2V and V2I communication for such as tunnels or carparks where the outdoor limitations of
localization, in [73] a localization technique utilizing radio VLC do not apply as a dense LED deployment can be real-
frequency identification (RFID) tags and readers, IMUs, and ized with constant uptime and with little ambient lighting,
V2V communication was proposed. The proposed method thereby providing low cost V2X communication with highly
used a matrix-based tag deployment strategy, where tags were accurate localization capability. Overall, VLC is a promising
deployed in clusters, with multiple adjacent clusters forming communication alternative for V2X applications, but appli-
rows of tags. Each vehicle was then provided a map of the cations to vehicle localization currently suffer from lack of
locations of each RFID tag. The advantage of the proposed research.
method is that accurate localization could be achieved with
low-density and low-cost RFID infrastructure. As a first step,
the method was evaluated in a highway simulation with aver- C. Discussion
age vehicle speeds of 108 km/h. It was shown that using Management of network traffic and quality of service in
intertag intervals of 100 m, 500 m, and 900 m resulted in both V2V and V2I systems require signals with low latency
mean absolute errors of 0.5 m, 1.4 m, and 2.9 m, respectively. as low as 10 ms [4]. Packet loss within the network should
Next, the proposed method was evaluated in experiments be also minimized, therefore, the network parameters (e.g.,
with intertag intervals of 200 m and two vehicles driven signal bandwidth, transmission power, etc.) need to be opti-
at speeds of 60 km/h. Mean absolute errors of 0.71 m mized for scalability of the networks as interference in
for localization were achieved, with higher errors observed congested networks can cause high packet loss ratios and
when the vehicle was turning. Therefore, even with relatively reduce the effective reliable transmission range by up to
sparsely deployed tags, resulting in a cost of a few dollars per 90% [86]. Moreover, in [51], it was noted that while coop-
kilometer of road, submeter localization accuracy could be erative localizations can perform effectively under high speed
achieved. and good communication conditions, in poor communication
Another emergent communication technology is visible conditions the effectiveness can be significantly reduced. The
light communication (VLC), which has gained interest for performance degradations can be overcome by modifying the
V2I [74]–[76] and V2V [77]–[80] applications. VLC utilizes communication parameters such as transmission power, packet
low cost light-emitting diodes (LEDs) to transmit informa- transmission frequency, and data rate. However, this can intro-
tion by turning the LED light on and off at a rate not duce other problems such as channel congestion. A further
TABLE III
S UMMARY OF V2I L OCALIZATION T ECHNIQUES
limitation of the V2V communication systems is discussed a security mechanism for the network will be required to
in [87], where the applicability of various routing protocols protect the integrity of the network whilst maintaining the
is discussed for VANET applications. The node behavior in users’ privacy [88].
urban and highway environments can vastly differ due to A summary of V2I localization techniques is represented
differences in vehicle speed, mobility patterns, and vehicle in Table III. The comparison of the results in Tables I–III
densities. This means that different routing protocols will have shows that the accuracy of localization systems highly depends
varying levels of performance in urban and highway scenarios; on the information sources utilized. For instance, among the
while some are more suited to the urban environments others sensor-based techniques, LiDAR-based techniques can achieve
are more suited to highway environments. Therefore, a hybrid very high accuracy and robustness at the cost of high power
routing protocol may be required that can adapt itself to dif- requirement and costly sensor with very limited performance
ferent environments. Furthermore, the security of the network in harsh environment conditions. It was also shown that the
must also be ensured. Falsified or altered messages or denial- methods such as camera-based localization or LGPR tech-
of-service attacks could severely impact the safety of VANET niques can offer adequate accuracy for autonomous vehicles at
applications. Similarly, the privacy of the network needs to be a lower cost at the expense of the lack of robustness against
maintained since most drivers will want to maintain the privacy variation of environmental conditions. In general, the main
and protection of any information sent over the network to limitations of all sensor-based techniques are their limited
avoid the possibility of their position being tracked. Therefore, line of sight, out of range environmental information, and
TABLE IV
L OCALIZATION T ECHNIQUES A DEQUATE FOR AUTONOMOUS V EHICLES localization of autonomous applications; however, the high
power and processing requirements and its high cost render
it unfeasible from cost-efficiency and commercialization point
of view. Therefore, further optimization of LiDAR technology
or alternative approaches such as localizing ground penetrating
Radar or vision-based localization within LiDAR maps could
offer a path toward commercially feasible systems. However,
further field testing to validate the robustness of these systems,
validating their performance under a variety of driving con-
ditions, and refining operation parameters will be necessary
before mass production.
It was also highlighted that integrating V2V information
with on-board sensory information has great potential to
improve the accuracy, robustness, and reliability of senor-
based localization systems. The main advantages of the coop-
erative systems are their capability of collecting information
out of line-of-sight and wider range, robustness to environ-
mental conditions, and low cost. However, V2V techniques
are limited by the shared position estimates and the number
limited operation in harsh environment. V2V techniques have of connected vehicles in the area. Infrastructure with known
presented great potential to address such limitations of sensor- position can be used to share accurate position estimates and
based techniques. Currently, the only available V2V local- with well-planned implementation, an adequate number of
ization technique is multilateration which can significantly position estimates can always be available. However, many of
improve the accuracy of GPS/IMU sensor-based technique the current cooperative systems have been evaluated through
when an adequate number of connected vehicles are avail- simulation, and while this is useful for the feasibility study
able. For example, the weighted average localization in [58] and initial performance evaluation, a variety of field tests will
achieved very high accuracy in a network of 200 cars, but the be required under different possible driving and communica-
accuracy decreased as the size of the networks decreased. The tion conditions before these systems are ready to be deployed.
VANET quality of service also limits the reliability of V2V Moreover, the quality of service in V2V and V2I networks
techniques since network noise will affect the received signals must be optimized for all possible operating conditions to
causing erroneous inputs into the localization algorithm. On ensure efficient and effective operation once deployed. The
the other hand, V2I localization techniques can mitigate the large scale of these networks means that the effect of char-
difficulty of guaranteeing an adequate number of signals and acteristics such as transmission power, packet transmission
accuracy of broadcasted position estimates in V2V localization frequency, data rate, RSU configuration, and network size must
techniques as the RSUs are installed in known fixed positions be studied and optimized to minimize latency and packet loss.
and the RSU density can be optimized to provide adequate Furthermore, the security of these networks must be consid-
accuracy and robustness. However, V2I techniques require ered to ensure the safety and privacy of all communication
costly infrastructure implementations to ensure high accu- over the network, as well as the effect of any security mech-
racy and robustness. Alternatively, utilizing existing cellular anisms to the performance of the network. Also, the effect of
networks provides a method for V2I localization without the network failure on the performance of the autonomous vehicles
need for implementation of new costly infrastructure. Similar must be considered. Furthermore, the emergent 5G communi-
to V2V techniques, the quality of service in V2I networks is cation systems could facilitate secure, reliable, and ultralow
also a limitation for implementation as noise can affect the latency communications among vehicles and with infrastruc-
received signals causing erroneous inputs and packet loss and ture, assisting cooperative localization techniques suitable for
latency can cause degradation in performance or failure of autonomous driving.
localization systems. A summary of localization techniques The environment in which the solution is to be deployed
considered adequate for autonomous vehicle localization from must be considered as well. As previously discussed, each of
a performance point of view is given in Table IV. the sensors and localization techniques have distinct pros and
cons. Therefore, the best solution may depend on environ-
mental factors such as weather, traffic, or road conditions. For
V. C ONCLUSION instance, camera-based solutions which depend on road mark-
In this paper, a concise survey of the current state-of-the- ing detection would not be feasible in parts of the world where
art of vehicle localization techniques using on-board sensing road markings are covered, e.g., by snow most of the year
systems and their combinations with V2V and V2I systems or on roads which lack these markings completely. Similarly,
was provided along with potentials and limitations of each V2X solutions rely heavily on dense networks of vehicles
technique on the implementation on autonomous vehicles. and/or infrastructure. While dense infrastructure deployment
It was shown that, from the performance point of view, and a large number of nearby vehicles may be a reasonable
the LiDAR techniques show the greatest promise for the assumption, for example, in urban or highway environments,
this might not be the case for rural environments. Also, 5G Fifth generation.
techniques which rely on GPS have to account for environ- AGPS Assisted GPS.
ments where GPS signals could be completely unavailable, AOA Angle-of-arrival.
such as tunnels. In such environments V2I solutions such as CAM Cooperative awareness message.
VLC may be required to provide reliable localization. COVEL Cooperative vehicle localization for efficient
Further research will be required to validate and refine urban mobility.
the performance and robustness of the most promising local- DENM Decentralised environmental notification
ization techniques. Moreover, the integration of cooperative message.
techniques with on-board sensing system will need further DGPS Differential global positioning system.
research to find the optimal way to fuse these localization DSRC Dedicated short range communication.
techniques together for increased reliability and accuracy. ECEF Earth centered earth fixed.
Furthermore, the optimal configuration of vehicular ad hoc ECI Earth-centered inertial.
networks will need to be investigated to achieve low packet EKF Extended Kalman filter.
loss and latency in highly congested networks with high-speed FMCW Frequency modulation continuous wave.
nodes and varying topology. Additionally, the reliability of GIS Geographical information system.
localization systems needs to be improved before commer- GNSS Global navigation satellite systems.
cialization, including consideration of sensor failures and fault GPS Global positioning system.
tolerant systems. GSM Global system for mobile communications.
HD High definition.
A PPENDIX A
HMI Human–machine interface.
D EFINITIONS
IEEE Institute of electrical and electronics engi-
Here is a list of definitions of terms used in this report. neers.
1) Accuracy: How close the measured position or value is IMU Inertial measurement unit.
from the systems true position or value. IOT Internet of Things.
2) Dead Reckoning: A localization method using an initial IOV Internet of Vehicles.
known position and odometry to calculate the changing IR-UWB Impulse radio ultrawide bandwidth.
position during movement. ITS Intelligent transportation system.
3) Error: The difference between an estimated value or KF Kalman filter.
position to the true value or position. LDM Local dynamic map.
4) Feature Map: An information layer which includes LED Light-emitting diode.
the environment features (e.g., road marks, buildings, LGPR Localizing ground penetrating radar.
obstacles, other road users, etc.) and their locations. LiDAR Light detection and ranging.
5) IEEE 802.11p: An addition to the IEEE 802.11 stan- LRF Laser range finder.
dard to include wireless access in vehicular environ- MAC Media access control.
ments communication system and defines modifications MAFS Master antenna fixed stations.
required to support it. IEEE 802.11 standard defines the MANET Mobile ad hoc network.
specifications for Wi-Fi products. MSE Mean square error.
6) Internet of Things: A network of interconnected tech- PF Particle filter.
nologies, devices, objects, and services. See [5] for PHY Physical layer.
further details. Radar Radio detecting and ranging.
7) Odometry: The measure of change in position over time, RANSAC Random sample consensus.
relative to the starting position. RFID Radio frequency identification.
8) Packet Loss: The loss of packets, units of information RMS Root mean square.
transferred over networks, as they fail to reach their RSS Radio signal strength.
destination. Typically caused by network congestion. RSU Road side unit.
9) Robustness: Ability of the system to cope with errors RTK Real time kinematic.
and erroneous inputs. SAM Service announcement message.
10) VANET: Vehicular ad hoc networks (VANETs) are a type SLAM Simultaneous localization and mapping.
of mobile ad hoc network, in which each node is free
SNR Signal-to-noise-ratio.
to move independently and will therefore frequently
SRR Short range Radar.
change which nodes they are connected to. VANETs are
TDOA Time-difference-of-arrival.
expected to use the IEEE 802.11p standard for wireless
TOA Time-of-arrival.
communication.
TTFF Time-to-first-fix.
UHF Ultrahigh frequency.
A PPENDIX B UWB Ultrawide band.
L IST OF A BBREVIATIONS V2I Vehicle-to-infrastructure.
Abbreviation Description V2V Vehicle-to-vehicle.
3-D Three dimensional. V2X Vehicle-to-everything.
VANET Vehicular ad hoc network. [21] J. Levinson, M. Montemerlo, and S. Thrun, “Map-based precision
VLC Visible light communication. vehicle localization in urban environments,” in Robotics: Science and
Systems, vol. 3. Cambridge, MA, USA: MIT Press, 2007, pp. 121–128.
WAVE Wireless access in vehicular environment. [22] G. M. Djuknic and R. E. Richton, “Geolocation and assisted GPS,”
Computer, vol. 34, no. 2, pp. 123–125, Feb. 2001.
[23] European Commission. (2017). Galileo. Accessed: Mar. 15, 2017.
R EFERENCES [Online]. Available: https://ec.europa.eu/growth/sectors/space/galileo_en
[24] European Commission. (2011). Satellite Navigation—Why Galileo?
[1] Bureau of Transportation Statistics. (2016). Table 1-11: Number of U.S. Accessed: Mar. 15, 2017. [Online]. Available: https://ec.europa.eu/
Aircraft, Vehicles, Vessels, and Other Conveyances. Accessed: growth/sectors/space/galileo_en
Dec. 1, 2016. [Online]. Available: http://www.rita.dot.gov/bts/ [25] NavStar. (1996). NavStar GPS User Equipment Introduction. Accessed:
sites/rita.dot.gov.bts/files/publications/national_transportation_statistics/ May 16, 2017. [Online]. Available: https://www.navcen.uscg.gov/pubs/
html/table_01_11.html gps/gpsuser/gpsuser.pdf
[2] Department for Transport. (May 2016). Research on the Impacts [26] F. Zhang et al., “A sensor fusion approach for localization with cumula-
of Connected and Autonomous Vehicles (CAVs) on Traffic Flow: tive error elimination,” in Proc. IEEE Conf. Multisensor Fusion Integr.
Summary Report. Accessed: Mar. 29, 2017. [Online]. Available: https:// Intell. Syst. (MFI), 2012, pp. 1–6.
www.gov.uk/government/uploads/system/uploads/attachment_data/file/ [27] C. Li, B. Dai, and T. Wu, “Vision-based precision vehicle localiza-
530091/impacts-of-connected-and-autonomous-vehicles-on-traffic-flow- tion in urban environments,” in Proc. Chin. Autom. Congr. (CAC),
summary-report.pdf Changsha, China, 2013, pp. 599–604.
[3] K. Jo, J. Kim, D. Kim, C. Jang, and M. Sunwoo, “Development of [28] I. Parra, M. A. Sotelo, D. Llorca, and M. Ocaña, “Robust visual odom-
autonomous car—Part I: Distributed system architecture and devel- etry for vehicle localization in urban environments,” Robotica, vol. 28,
opment process,” IEEE Trans. Ind. Electron., vol. 61, no. 12, no. 3, pp. 441–452, 2010.
pp. 7131–7140, Dec. 2014. [29] S. Kamijo, Y. Gu, and L. Hsu, “Autonomous vehicle technologies:
[4] 5G-PPP. (2015). 5G Automotive Vision. Accessed: Feb. 12, 2017. Localization and mapping,” IEICE Fundam. Rev., vol. 9, no. 2,
[Online]. Available: https://5g-ppp.eu/wp-content/uploads/2014/02/5G- pp. 131–141, 2015.
PPP-White-Paper-on-Automotive-Vertical-Sectors.pdf [30] J. K. Suhr, J. Jang, D. Min, and H. G. Jung, “Sensor fusion-based
[5] R. Minerva, A. Biru, and D. Rotondi. (2015). Towards a Definition of low-cost vehicle localization system for complex urban environments,”
the Internet of Things (IoT). [Online]. Available: https://iot.ieee.org/ IEEE Trans. Intell. Transp. Syst., vol. 18, no. 5, pp. 1078–1086,
images/files/pdf/IEEE_IoT_Towards_Definition_Internet_of_Things_ May 2017.
Revision1_27MAY15.pdf [31] E. Ward and J. Folkesson, “Vehicle localization with low cost
[6] Y. Fangchun, W. Shangguang, L. Jinglin, L. Zhihan, and S. Qibo, “An radar sensors,” in Proc. IEEE Intell. Veh. Symp. (IV), 2016,
overview of Internet of Vehicles,” China Commun., vol. 11, no. 10, pp. 864–870.
pp. 1–15, Oct. 2014. [32] D. Vivet, F. Gérossier, P. Checchin, L. Trassoudaine, and R. Chapuis,
[7] Asia-Pacific Economic Cooperation (APEC), “White paper of Internet “Mobile ground-based radar sensor for localization and mapping: An
of Vehicles,” in Proc. 50th Telecommun. Inf. Working Group Meeting, evaluation of two approaches,” Int. J. Adv. Robot. Syst., vol. 10, no. 5,
Brisbane, QLD, Australia, 2014. pp. 307–318, 2013.
[8] Y. Cao and N. Wang, “Toward efficient electric-vehicle charging using [33] M. Cornick, J. Koechling, B. Stanley, and B. Zhang, “Localizing
VANET-based information dissemination,” IEEE Trans. Veh. Technol., ground penetrating RADAR: A step toward robust autonomous ground
vol. 66, no. 4, pp. 2886–2901, Apr. 2017. vehicle localization,” J. Field Robot., vol. 33, no. 1, pp. 82–102,
[9] Y. Cao et al., “A cost-efficient communication framework for battery- 2016.
switch-based electric vehicle charging,” IEEE Commun. Mag., vol. 5, [34] Velodyne LiDAR. (2016). Product Guide. Accessed: Nov. 28, 2016.
no. 5, pp. 162–169, May 2017. [Online]. Available: http://velodynelidar.com/docs/datasheet/LiDAR%20
[10] N. Lu, N. Cheng, N. Zhang, X. Shen, and J. W. Mark, “Connected Comparison%20chart_Rev-A_Web.pdf
vehicles: Solutions and challenges,” IEEE Internet Things J., vol. 1, [35] A. Y. Hata and D. F. Wolf, “Feature detection for vehicle localization
no. 4, pp. 289–299, Aug. 2014. in urban environments using a multilayer LIDAR,” IEEE Trans. Intell.
[11] Mobileye. (2017). The Mapping Challenge. Accessed: Mar. 29, 2017. Transp. Syst., vol. 17, no. 2, pp. 420–429, Feb. 2016.
[Online]. Available: http://www.mobileye.com/our-technology/rem/ [36] J. Levinson and S. Thrun, “Robust vehicle localization in urban environ-
[12] NVIDIA. (2017). Introducing the New Nvidia Drive px2. Accessed: ments using probabilistic maps,” in Proc. IEEE Int. Conf. Robot. Autom.
Apr. 20, 2017. [Online]. Available: http://www.nvidia.com/object/drive- (ICRA), 2010, pp. 4372–4378.
px.html [37] J. Castorena and S. Agarwal, “Ground-edge-based LIDAR localization
[13] OpenStreetMap. (2016). Micromapping. Accessed: Mar. 29, 2017. without a reflectivity calibration for autonomous driving,” IEEE Robot.
[Online]. Available: http://wiki.openstreetmap.org/wiki/Micromapping Autom. Lett., vol. 3, no. 1, pp. 344–351, Jan. 2018.
[14] ISO. (2015). ISO/TS 18750:2015, Intelligent Transport Systems— [38] H. Kim, B. Liu, C.-Y. Goh, S. Lee, and H. Myung, “Robust vehi-
Cooperative Systems—Definition of a Global Concept for Local cle localization using entropy-weighted particle filter-based data fusion
Dynamic Maps. Accessed: Mar. 29, 2017. [Online]. Available: of vertical and road intensity information for a large scale urban
https://www.iso.org/standard/63273.html area,” IEEE Robot. Autom. Lett., vol. 2, no. 3, pp. 1518–1524,
[15] ETSI. (2014). Intelligent Transport Systems (ITS); Vehicular Jul. 2017.
Communications; Basic Set of Applications; Local Dynamic [39] R. W. Wolcott and R. M. Eustice, “Robust LIDAR localization using
Map (LDM). Accessed: Mar. 29, 2017. [Online]. Available: multiresolution Gaussian mixture maps for autonomous driving,” Int. J.
http://www.etsi.org/deliver/etsi_en/302800_302899/302895/01.00.00_20/ Robot. Res., vol. 36, no. 3, pp. 292–319, 2017.
en_302895v010000a.pdf [40] R. W. Wolcott and R. M. Eustice, “Visual localization within LIDAR
[16] N. Mattern and G. Wanielik, “Vehicle localization in urban envi- maps for automated urban driving,” in Proc. IEEE Int. Conf. Intell.
ronments using feature maps and aerial images,” in Proc. 14th Int. Robots Syst., 2014, pp. 176–183.
IEEE Conf. Intell. Transp. Syst. (ITSC), Washington, DC, USA, 2011, [41] L. Wei, C. Cappelle, and Y. Ruichek, “Horizontal/vertical LRFs and GIS
pp. 1027–1032. maps aided vehicle localization in urban environment,” in Proc. IEEE
[17] G. Vivo. (2010). SAFESPOT Core Architecture—LDM API and Usage Annu. Conf. Intell. Transp. Syst. (ITSC), The Hague, The Netherlands,
Reference. [Online]. Available: http://www.safespot-eu.org/documents/ 2013, pp. 809–814.
SF_D7.3.1_Annex2_LDM_API_and_Usage_Reference_v0.7.pdf [42] S. Jung, J. Kim, and S. Kim, “Simultaneous localization and mapping of
[18] F. Nelli. (2016). 3D Laser Scanning. Accessed: Mar. 29, 2017. [Online]. a wheel-based autonomous vehicle with ultrasonic sensors,” Artif. Life
Available: http://www.meccanismocomplesso.org/en/laser-scanning-3d/ Robot., vol. 14, no. 2, pp. 186–190, 2009.
[19] I. Skog and P. Händel, “In-car positioning and navigation technologies— [43] M. Obst, N. Mattern, R. Schubert, and G. Wanielik, “Car-to-car com-
A survey,” IEEE Trans. Intell. Transp. Syst., vol. 10, no. 1, pp. 4–21, munication for accurate vehicle localization—The CoVeL approach,” in
Mar. 2009. Proc. 9th Int. Multi Conf. Syst. Signals Devices, 2012, pp. 1–6.
[20] H.-S. Tan and J. Huang, “DGPS-based vehicle-to-vehicle cooperative [44] K. Cheok, G. Smid, G. Hudas, and J. Overholt, An Ultra-Wideband RF
collision warning: Engineering feasibility viewpoints,” IEEE Trans. Method for Localizing an Autonomous Mobile Robot, Jadi Inc., Troy,
Intell. Transp. Syst., vol. 7, no. 4, pp. 415–428, Dec. 2006. MI, USA, 2004.
[45] M. Terré, M. Pischella, and E. Vivier, “UWB,” in Wireless [67] A. Varshavsky et al., “Are GSM phones THE solution for localiza-
Telecommunication Systems. London, U.K.: ISTE, 2013, pp. 155–163. tion?” in Proc. 7th IEEE Workshop Mobile Comput. Syst. Appl., 2006,
[46] F. Nekoogar, Ultra-Wideband Communications: Fundamentals and pp. 20–28.
Applications, 1st ed. Upper Saddle River, NJ, USA: Prentice-Hall, [68] M. Y. Chen et al., “Practical metropolitan-scale positioning for GSM
2006. phones,” in Proc. 8th UbiComb, Orange County, CA, USA, 2006,
[47] J. Ibanez-Guzman, S. Lefevre, A. Mokkadem, and S. Rodhaim, “Vehicle pp. 225–242.
to vehicle communications applied to road intersection safety, field [69] C.-H. Chen, C.-A. Lee, and C.-C. Lo, “Vehicle localization and veloc-
results,” in Proc. 13th Int. IEEE Annu. Conf. Intell. Transp. Syst., 2010, ity estimation based on mobile phone sensing,” IEEE Access, vol. 4,
pp. 192–197. pp. 803–817, 2016.
[48] P. Papadimitratos, A. de La Fortelle, K. Evenssen, R. Brignolo, and [70] H. Xue, H. Zhu, S. Cao, S. Chang, and J. Cao, “UPS: Combatting
S. Cosenza, “Vehicular communication systems: Enabling technologies, urban vehicle localization with cellular-aware trajectories,” in Proc.
applications, and future outlook on intelligent transportation,” IEEE Glob. Commun. Conf. (GLOBECOM), 2016, pp. 1–7.
Commun. Mag., vol. 47, no. 11, pp. 84–95, Nov. 2009. [71] L. C. Bento, R. Parafita, and U. Nunes, “Inter-vehicle sensor fusion
[49] Intelligent Transport Systems (ITS); Vehicular Communications; Basic for accurate vehicle localization supported by V2V and V2I commu-
Set of Applications; Part 2: Specification of Cooperative Awareness nications,” in Proc. 15th Int. IEEE Conf. Intell. Transp. Syst. (ITSC),
Basic Service, ETSI Standard EN 302 637-2, 2014. [Online]. Available: Anchorage, AK, USA, 2012, pp. 907–914.
http://www.etsi.org/deliver/etsi_en/302600_302699/30263702/01.03.02_ [72] L. Bento, R. Chelim, and U. Nunes, “Collaborative vehicle self-
60/en_30263702v010302p.pdf localization using multi-GNSS receivers and V2V/V2I communica-
[50] Intelligent Transport Systems (ITS); Vehicular Communications; tions,” in Proc. IEEE 18th Int. Conf. Intell. Transp. Syst. (ITSC), 2015,
Basic Set of Applications; Part 3: Specifications of Decentralized pp. 2525–2532.
Environmental Notification Basic Service, ETSI Standard EN 302 [73] H. Qin, Y. Peng, and W. Zhang, “Vehicles on RFID: Error-cognitive
637-3, 2014. [Online]. Available: http://www.etsi.org/deliver/etsi_en/ vehicle localization in GPS-less environments,” IEEE Trans. Veh.
302600_302699/30263703/01.02.01_30/en_30263703v010201v.pdf Technol., vol. 66, no. 11, pp. 9943–9957, Nov. 2017.
[51] M. Sepulcre, J. Gozalvez, and J. Hernandez, “Cooperative vehicle-to- [74] S. Arai et al., “Experimental on hierarchical transmission scheme for
vehicle active safety testing under challenging conditions,” Transp. Res. visible light communication using LED traffic light and high-speed cam-
C Emerg. Technol., vol. 26, pp. 233–255, Jan. 2013. era,” in Proc. IEEE 66th Veh. Technol. Conf. (VTC Fall), Baltimore, MD,
[52] S. Fujii et al., “Cooperative vehicle positioning via V2V communications USA, 2007, pp. 2174–2178.
and onboard sensors,” in Proc. Veh. Technol. Conf., 2011, pp. 1–5. [75] S. Arai et al., “Multiple LED arrays acquisition for image-sensor-based
[53] M. Rohani, D. Gingras, V. Vigneron, and D. Gruyer, “A new decen- I2V-VLC using block matching,” in Proc. IEEE 11th Consum. Commun.
tralized Bayesian approach for cooperative vehicle localization based Netw. Conf. (CCNC), Las Vegas, NV, USA, 2014, pp. 605–610.
on fusion of GPS and VANET based inter-vehicle distance measure- [76] C. Li and S. Shimamoto, “An open traffic light control model for reduc-
ment,” IEEE Intell. Transp. Syst. Mag., vol. 7, no. 2, pp. 85–95, ing vehicles’ CO2 emissions based on ETC vehicles,” IEEE Trans. Veh.
Apr. 2015. Technol., vol. 61, no. 1, pp. 97–110, Jan. 2012.
[54] N. Mattern, M. Obst, R. Schubert, and G. Wanielik, “Co-operative vehi- [77] C. B. Liu, B. Sadeghi, and E. W. Knightly, “Enabling vehicular visible
cle localization algorithm—Evaluation of the COVEL approach,” in light communication (V2LC) networks,” in Proc. 8th ACM Int. Workshop
Proc. 9th Int. Multi Conf. Syst. Signals Devices, 2012, pp. 1–5. Veh. Inter Netw., Las Vegas, NV, USA, 2011, pp. 41–50.
[78] L.-C. Wu and H.-M. Tsai, “Modeling vehicle-to-vehicle visible light
[55] R. Ordóñez-Hurtado, W. Griggs, E. Crisostomi, and R. Shorten,
communication link duration with empirical data,” in Proc. IEEE
“Cooperative positioning in vehicular ad-hoc networks supported
Globecom Workshops, Atlanta, GA, USA, 2013, pp. 1103–1109.
by stationary vehicles,” Feb. 2015. [Online]. Available: https://
[79] S.-H. You, S.-H. Chang, H.-M. Lin, and H.-M. Tsai, “Visible light com-
www.hamilton.ie/smarttransport/publications/arXivCooperativePosition
munications for scooter safety,” in Proc. MobiSys, Taipei, Taiwan, 2013,
ingFeb2015.pdf
pp. 509–510.
[56] K. Golestan, S. Seifzadeh, M. Kamel, F. Karray, and F. Sattar, “Vehicle
[80] S.-H. Yu, O. Shih, H.-M. Tsai, N. Wisitpongphan, and R. D. Roberts,
localization in VANETs using data fusion and V2V communication,” in
“Smart automotive lighting for vehicle safety,” IEEE Commun. Mag.,
Proc. 2nd ACM Int. Symp. Design Anal. Intell. Veh. Netw. Appl., 2012,
vol. 51, no. 12, pp. 50–59, Dec. 2013.
pp. 123–130.
[81] R. Roberts, P. Gopalakrishnan, and S. Rathi, “Visible light position-
[57] F. Ahammed, J. Taheri, A. Y. Zomaya, and M. Ott, International ing: Automotive use case,” in Proc. IEEE Veh. Netw. Conf. (VNC),
Conference on Mobile and Ubiquitous Systems: Computing, Networking, Jersey City, NJ, USA, 2010, pp. 309–314.
and Services. Berlin, Germany: Springer, 2010. [82] S. Rajagopal, R. D. Roberts, and S.-K. Lim, “IEEE 802.15.7 visible
[58] L. Altoaimy and I. Mahgoub, “Fuzzy logic based localization for vehic- light communication: Modulation schemes and dimming support,” IEEE
ular ad hoc networks,” in Proc. IEEE Symp. Comput. Intell. Veh. Transp. Commun. Mag., vol. 50, no. 3, pp. 72–82, Mar. 2012.
Syst. (CIVTS), 2014, pp. 121–128. [83] P. H. Pathak, X. Feng, P. Hu, and P. Mohapatra, “Visible light commu-
[59] A. Boukerche, H. Oliveira, E. Nakamura, and A. Loureiro, “Vehicular ad nication, networking, and sensing: A survey, potential and challenges,”
hoc networks: A new challenge for localization-based systems,” Comput. IEEE Commun. Surveys Tuts., vol. 17, no. 4, pp. 2047–2077, 4th Quart.,
Commun., vol. 31, no. 12, pp. 2838–2849, 2008. 2015.
[60] J. Vales-Alonso, F. Vicente-Carrasco, and J. J. Alcaraz, “Optimal con- [84] L. Li, P. Hu, C. Peng, G. Shen, and F. Zhao, “Epsilon: A visible light
figuration of roadside beacons in V2I communication,” Comput. Netw., based positioning system,” in Proc. 11th USENIX Symp. Netw. Syst.
vol. 55, no. 14, pp. 3142–3153, 2011. Design Implement. (NSDI), Seattle, WA, USA, 2014, pp. 331–343.
[61] O. Hassan, I. Adly, and K. A. Shehata, “Vehicle localization system [85] B. W. Kim and S.-Y. Jung, “Vehicle positioning scheme using V2V and
based on IR-UWB for V2I applications,” in Proc. 8th Int. Conf. Comput. V2I visible light communications,” in Proc. IEEE 83rd Veh. Technol.
Eng. Syst. (ICCES), Cairo, Egypt, 2013, pp. 133–137. Conf. (VTC Spring), Nanjing, China, 2016, pp. 1–5.
[62] A. Fascista, G. Ciccarese, A. Coluccia, and G. Ricci, “A localization [86] R. Schmidt, T. Köllmer, T. Leinmüller, B. Böddeker, and
algorithm based on V2I communications and AOA estimation,” IEEE G. Schäfer, “Degradation of Transmission Range in VANETs
Signal Process. Lett., vol. 24, no. 1, pp. 126–130, Jan. 2017. caused by Interference,” Praxis Der Informationsverarbeitung Und
[63] A. Khattab, Y. A. Fahmy, and A. A. Wahab, “High accuracy GPS-free Kommunikation, vol. 32, no. 4, pp. 224–234, 2009.
vehicle localization framework via an INS-assisted single RSU,” Int. J. [87] A. Fonseca and T. Vazão, “Applicability of position-based routing for
Distrib. Sensor Netw., vol. 11, no. 5, 2015, Art. no. 795036. VANET in highways and urban environment,” J. Netw. Comput. Appl.,
[64] N. Houdali, T. Ditchi, E. Gérome, J. Lucas, and S. Holé, “RF infrastruc- vol. 36, no. 3, pp. 961–973, 2013.
ture cooperative systems for in lane vehicle localization,” Electronics, [88] R. G. Engoulou, M. Bellaïche, S. Pierre, and A. Quintero, “VANET
vol. 3, no. 4, pp. 598–608, 2014. security surveys,” Comput. Commun., vol. 44, no. 5, pp. 1–13, 2014.
[65] J. A. del Peral-Rosado, J. A. López-Salcedo, S. Kim, and
G. Seco-Granados, “Feasibility study of 5G-based localization for
assisted driving,” in Proc. Int. Conf. Localization GNSS (ICL GNSS),
2016, pp. 1–6.
[66] S. Ergen et al., “RSSI-fingerprinting-based mobile phone localization
with route constraints,” IEEE Trans. Veh. Technol., vol. 63, no. 1,
pp. 423–428, Jan. 2014. Authors’ photographs and biographies not available at the time of publication.

A Survey of The State-of-the-Art Localization Techniques and Their Potentials For Autonomous Vehicle Applications

Uploaded by

Copyright:

Available Formats

A Survey of The State-of-the-Art Localization Techniques and Their Potentials For Autonomous Vehicle Applications

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Survey of The State-of-the-Art Localization Techniques and Their Potentials For Autonomous Vehicle Applications

Uploaded by

Copyright:

Available Formats

IEEE INTERNET OF THINGS JOURNAL, VOL. 5, NO.

2, APRIL 2018 829

A Survey of the State-of-the-Art Localization

Fig. 1. Local dynamic map layers with example information. Adapted

Fig. 3. Differential GPS correction. Gray objects indicate estimated position

coordinate system such as Earth-centered inertial (ECI) coor-

on static maps, which meant that extreme changes to the

A. Vehicle-to-Vehicle Localization Techniques

AOA measurements to improve localization accuracy. Also,

You might also like