Intelligent Parking Lot Application Using Wireless Sensor Networks

Sangwon Lee, Dukhee Yoon, Amitabha Ghosh

Autonomous Networks Research Group
Ming Hsieh Department of Electrical Engineering
University of Southern California, Los Angeles, CA 90089
{sangwonl, dukheeyo, amitabhg}@usc.edu

ABSTRACT With the increasing growth of automotive industry, the de-

mand for intelligent parking service is expected to grow
Wireless sensor networks offer an attractive choice for low- rapidly in the near future. This emerging service will pro-
cost and easy-to-deploy solutions for intelligent traffic guid- vide automatic management of parking lots by accurate
ance systems and parking lot applications. In this paper, monitoring and making that information available to cus-
we propose the use of a combination of magnetic and ultra- tomers and facility administrators. Wireless sensor net-
sonic sensors for accurate and reliable detection of vehicles works have a great potential toward providing a cost effec-
in a parking lot. We describe a modified version of the min- tive solution to this service for a variety of reasons, such
max algorithm [10] for detection of vehicles using mag- as their ease of deployment in existing parking lots without
netometers, and also an algorithm for ultrasonic sensors. having to install new, expensive cabling, and the flexibility
Through extensive real world experiments conducted in a to couple them with sophisticated but cheap sensors with
multi-storied university parking space we compare the pros different modalities that can accurately keep track of ve-
and cons of using different sensing modalities, and show hicles. Moreover, information gathered by each node can
that ultrasonic sensors along with magnetometers is an ex- be collaboratively processed in a distributed or centralized
cellent choice for accurate vehicle detections. We demon- way to evaluate other meaningful metrics such as duration
strate the efficacy of our proposed approach by conducting of parking, automatic billing and payment, etc., to the ben-
an elaborate car counting experiment lasting over a day, efit of users and administrators.
and show promising results using these two sensing modal-
ities.
A few existing solutions focus on parking lot applica-
tions using sensor technologies, such as magnetometers and
KEYWORDS: wireless sensor networks, traffic guid- video cameras. However, magnetometers are very sensi-
ance systems, ultrasonic sensors, magnetometers tive on environmental factors, as a result of which their de-
tections are not always accurate. Moreover, since magne-
tometers measure the change in magnitude and direction of
1. INTRODUCTION Earth’s magnetic field caused by the presence of a vehicle,
they need to be placed at close proximity to the vehicle. Al-
In recent years, wireless sensor networks (WSN) [1] have though this might be possible near the entrance of a parking
inspired tremendous research interest in diverse applica- lot, it is very difficult to place them in close proximity to ve-
tion domains such as structural health monitoring [23], un- hicles on upper floors simply because there are typically no
derwater marine life monitoring [2], military and security entrance marked for upper floors and vehicles move at rela-
surveillance, health-care, smart homes, automotive indus- tively higher speeds than near the entrance. On the other
try, etc. A typical sensor network consists of hundreds hand, video camera based solutions are energetically ex-
to thousands of sensor nodes, each equipped with various pensive and they can generate large amount of data which
kinds of sensors, deployed over a geographical region of could be very difficult to transmit over multiple hops in a
interest. A sensor node by itself has severe resource con- wireless environment. These disadvantages coupled with
straints in terms of memory, battery power, computation the fact that there are other objects moving in a parking lot,
and communication capabilities; however, a group of sen- such as humans, greatly reduce the applicability of only one
sors collaborating with each other can accomplish a much type of sensor technology, i.e., only magnetometers or only
bigger task efficiently. video cameras for cheap and accurate parking lot manage-

978-1-4244-2249-4/08/$25.00 ©2008 IEEE 48

Authorized licensed use limited to: University of Southern California. Downloaded on November 25, 2009 at 15:56 from IEEE Xplore. Restrictions apply.
 ment solutions. especially temperature and acoustic sensors might be use-
less. Light sensors could work but might malfunction under
In this paper, we propose a hybrid approach for an intel- shadows. Moreover, it is hard to manage a lot of sensor
ligent parking system using a combination of ultrasonic nodes and the overall system could be more expensive.
and magnetic sensors, and demonstrate promising results
through real world experiments performed at our multi- A traffic surveillance system is described in [7], which
storied university parking structure. Because of practical uses magnetometers to detect the presence and estimate
limitations, such as hardware characteristics and localiza- the speed of vehicles near street intersections and parking
tion errors, we show that hybrid solutions using combina- lots. Using magnetic signatures the system could also clas-
tion of sensing technologies are more practical and accu- sify and re-identify vehicles. MIT Intelligent Transporta-
rate. Unlike some of the previous works [20] that propose to tion System [15] and Responsive Roadways [14], [18] are
place a magnetic sensor at each parking space and thereby also examples of transportation applications using wireless
provide occupancy information for each space at the cost of magnetic sensors. However, in all these systems the sensor
expensive infrastructure, our goal is to count the number of nodes need to be glued on the pavement or placed just un-
vehicles only on each floor and provide a cheap but accurate der the road surface where vehicles are to be detected. This
solution. is particularly not suitable for parking lots because vehicles
could move anywhere inside a parking lot unlike roads, and
The rest of the paper is organized as follows. In Section 2, thus could possibly damage sensors.
we discuss related works. In Section 3, we first describe
the devices that are used in our implementation and then In [5], design and implementation issues for a reliable WSN
describe our sensing technique. Section 4 details on the de- system using magnetic sensors are presented, which can
tection algorithms for magnetic and ultrasonic sensors. We track available parking spaces in real time and commu-
present our detailed experimental results in Section 5, and nicate that information to the users. They propose a de-
finally we conclude in Section 6. tection scheme based on magnetometer signature measure-
ments and implement it on Mica2 motes [8]. In [4], and as
2. RELATED WORKS part of the D-Systems Project [3], an implementation of a
car-park management system using a tiered architecture is
In this section we review the existing literature on traffic provided using the DSYS25z [17] magnetic sensor boards
management systems, and in particular, parking lot applica- developed by Tyndall [22]. It highlights several problems
tions using wireless sensor networks. in wireless communications in a parking lot environment
and proposes recommendations such as use of dynamic and
The Parking Space Finder Application that is part of Iris- robust routing, delayed retransmissions, etc., based on ex-
Net [13] proposed a wide-area sensor network architecture, perimental results to overcome those problems.
in which video cameras, microphones, and motion detec-
tors are used to detect occupancy or availability of parking Our work most closely resembles in spirit to the Siemens
spaces. Using web technologies, users can acquire the pro- Sipark PMA [19] solution, in terms of both using ultrasonic
cessed information that is published on the web and gener- sensors to detect vehicles. Sipark PMA is perhaps the most
ated by feeding all these sensor-data. However, as men- prominent parking guidance system for multi-storied park-
tioned earlier, video cameras generate a large amount of ing lots where ultrasonic sensors are connected using a bus-
data that incur high energy expenditure and communication style network to distribute power and transmit sensing re-
bandwidth, both of which are limited in sensor networks. ports. When a car enters a parking lot, the guidance system
directs the driver to a free parking space along the shortest
In [20], an intelligent car parking system using wireless possible route. Our work differs from the existing literature
sensor networks is proposed, where each parking space is and compliments it in the following ways: (1) we propose
equipped with one sensor to detect its occupancy. The pro- a reliable vehicle detection algorithm using ultrasonic sen-
totype implementation is based on remote controlled toy sors, (2) we propose to use a combination of ultrasonic and
cars and uses Crossbow motes [8] that have light, temper- magnetic sensors to increase the accuracy of detection and
ature, and acoustic sensors. However, it is not mentioned minimize environmental effects, and show that other forms
how to detect a car with those sensors. It is argued that the of sensing such as light, temperature, infrared, and acoustic
proposed three layered system architecture will reduce the do not give accurate results, and (3) we conduct real world
cost of manpower and minimize human operations, how- extensive experiments to demonstrate the efficacy of our ap-
ever, as we show in this paper through real world experi- proach using these sensors mounted on inexpensive motes
ments that it is hard to even detect cars with those sensors; running TinyOS 2.x [21].

49

Authorized licensed use limited to: University of Southern California. Downloaded on November 25, 2009 at 15:56 from IEEE Xplore. Restrictions apply.
 (a) (b)
(a) (b)
Figure 1. (a) SBT80 Sensor Board, (b) SBT80 Sensor
Board Mounted on Tmote Sky. Figure 2. (a) Devantech SRF02 Ultrasonic Sensor, (b)
SRF02 Installed on Tmote Sky.
3. PROPOSED APPROACH
Figure 2(a) shows an SRF02 sensor while Table 2 highlights
In this section, we first describe the devices that are used for its properties. Figure 2(b) shows an SRF02 sensor attached
our experiments, and then describe the sensing technique to a Tmote Sky.
upon which we base our detection algorithms.
Table 2. SRF02 Ultrasonic Sensor Properties
3.1. Device Specifications
Protocol I2C, Serial
We use Tmote Sky manufactured by Moteiv [16] (now Sen-
Range 15cm - 6m
tilla) as the wireless nodes. Tmote Sky’s are IEEE WPAN
Response Time 70ms
standard 802.15.4 [12] compliant devices that are ideal for
Frequency 40kHz
mesh networking, featuring a 250 kbps radio, 10 kB RAM,
48 kB flash, and 1 MB storage. One mote is used as the Voltage 5v
base station and several others are used for measurements. Current 4mA
We equip each mote except for the base station with a multi-
modality sensor board, SBT80 from EasySen [11] that has
six different sensors, e.g., visual light, infrared, tempera-
ture, acoustic, magnetometer, and accelerometer, as illus- 3.2. Sensing Technique
trated in Figure 1(a) and Table 1. Figure 1(b) shows a Tmote Sensing technique is one of the most important concerns in
Sky mounted with an SBT board. order to accomplish an accurate and energy-efficient detec-
tion scheme. The sensing period should be short enough
Table 1. Sensors on SBT80 Board (e.g., 0.1 second) so as to not miss any moving vehicles;
however, too frequent sensing will consume lots of energy.
Sensor Type Manufacturer, Model
Therefore, there is a trade-off between the accuracy of mea-
Visual Light PerkinElmer, VTB9412B
surements and energy consumption. It is thus necessary to
Infrared Advanced Photonix, Inc., PDB-C139F
find an optimal sensing period. We do this by conducting
Acoustic Horn, EM6050
several experiments from analyzing the change in magnetic
Temperature Maxim Integrated Products, wave patterns obtained from a passing vehicle. This optimal
MAX6612MXK period is then fed in the main detection algorithm.
Magnetometer Honeywell, HMC1052
Accelerometer Freescale Semiconductor,
MMA6260Q 4. DETECTION ALGORITHMS
4.1. Modified Min-Max Algorithm For Magne-
tometers
For ultrasonic sensing we use the low cost Devantech We use a modified version of the min-max algorithm [10],
SRF02 [9] sensors. These sensors emit ultrasonic waves which is particularly suitable for analyzing the wave-
and perceive the reflected waves from an object to calculate formed raw magnetic and acoustic data. The overall struc-
its distance from the object. Thus, they provide much easier ture of the modified min-max algorithm is illustrated in Fig-
detection method than those based on acoustic, magnetic, or ure 3 through six basic steps. In the first step it obtains raw
visual light which are influenced by environmental factors. data from magnetometer measurements. Then it calculates

50

Authorized licensed use limited to: University of Southern California. Downloaded on November 25, 2009 at 15:56 from IEEE Xplore. Restrictions apply.
 track of the local minimums and maximums, defined by
Equations (2) and (3). Finally, the machine decides that a
vehicle has passed if it jumps from state hill count down
to f lat and the difference between the local maximum and
minimum is greater than a certain threshold T , determined
empirically as 70000.
⎧
⎪
⎪ min{f [n] − localM in[n − 1]},
⎪
⎪
⎨ if x ∈ {f lat, f lat count down}
localM in[n] = f [n], if x ∈ {hill count down} and
⎪
⎪
⎪
⎪ localM ax[n] − localM in[n] > T
⎩
unchanged, otherwise
(2)

⎧
⎨ max{f [n] − localM ax[n − 1]},
localM ax[n] = if x ∈ {hill count up}
⎩
unchanged, otherwise
(3)

4.2. Ultrasonic Sensor Detection Algorithm

Figure 3. Modified Min-max Detection Algorithm
The detection algorithm for ultrasonic sensors is rela-
tively simple and based on distance measurements from
a local weighted average by weighing the previous average the object to the sensors. A flowchart of the algo-
nine times of the new data. This step prevents oscillation rithm is shown in Figure 4. If the distance is smaller
of the local average by unwanted impulse data. In the third than a certain DIST AN CE T HRESHOLD, a counter
step, to increase the difference between the signal compo- is started, and incremented until the distance returns
nent and the noise component, a squared difference between to the normal value, at which point there is no ob-
the local average and the current raw data is calculated. As ject anymore. If the value of the counter is more than
we will see shortly, this step makes the signal component a certain DU RAT ION T HRESHOLD then a vehi-
more prominent and helps filtering. Step four applies a low- cle is detected. A moving person is easily distinguished
pass filter to reduce noise, which tends to be at a high fre- from a moving vehicle because a person’s passing du-
quency. Next, we calculate a moving average over forty ration is much smaller than that of a vehicle. The ex-
previous measurements. This choice of forty is made con- perimental results presented later demonstrate this fact.
sidering the limited memory and energy of Tmotes and the The values of the DIST AN CE T HRESHOLD and
accuracy of detection. In the last step, the moving average DU RAT ION T HRESHOLD depend on the placement
is fed in the min-max algorithm to detect the presence of a of the sensors in the parking lot and are empirically deter-
vehicle. mined to be 140 and 5, respectively, for our experiments.
The min-max detection algorithm has a state machine
which has five states, state(x) = {f lat, f lat count up, 5. EXPERIMENTAL RESULTS
f lat count down, hill count up, hill count down}. The
input to the state machine is the sign of the current slope 5.1. Sensing Experiments
and defined as:
⎧ The success of an intelligent parking application in accurate
⎨ sign(f [n] − f [n − 1]), if |f [n] − f [n − 1]| > and reliable detection of vehicles is crucially dependent on
u[n] = M IN DELT A U the types and locations of the sensors. To determine the best
⎩ sensing method for our vehicle counting experiment, we
0, otherwise
(1) first conduct several measurements using all the six sensors
where M IN DELT A U is a predefined positive constant, of the SBT80 multi-modality sensor board, the SRF02 ul-
2000. A counter, stateCounter is associated with each state, trasonic sensor, and visual light with laser pointer. We con-
which resets when the machine jumps to a new state, and ducted these experiments in a multi-storied parking struc-
counts up when the state loops back to itself. There are ture inside our university campus (see Figure 5(a)). Figure
also two variables, localM in[n] and localM ax[n], to keep 5(b) shows the placement of a mote mounted with a SBT80

51

Authorized licensed use limited to: University of Southern California. Downloaded on November 25, 2009 at 15:56 from IEEE Xplore. Restrictions apply.
 (a)

(b)

Figure 5. Test Bed: (a) Parking Structure Inside USC

Figure 4. Ultrasonic Sensor Detection Algorithm Campus. (b) A Tmote Mounted With SBT80 Placed
Near the Entrance.

sensor board near the entrance. The ultrasonic sensors are

passing cars. We use Java to test it on the laptop and port
attached to the ceiling.
the code to TinyOS programming language NesC. Figure 8
shows the graphs for each step of algorithm as described in
Figure 6 and 7 presents the results of one sample run of
Section 4.1. Clearly, even the raw data from magnetometer
our experiment with an optimal sensing period set to 70 ms.
y-axis show three fluctuating places, indicating three cars
The x-axis in all the graphs represent packet numbers and
have passed consecutively. In the next step, squaring of dif-
y-axis represents the measured value by each sensor in ap-
ference between the local average and the raw data makes
propriate units depending on the sensing modality. A car
the car-passing points more prominent than non-passing
enters the parking lot during the experiment corresponding
points. The low-pass filter eliminates the noise in the wave
to the packets from 100 to 130. Clearly, visual light and in-
forms, and the moving average reduces the fluctuating char-
frared sensors are not applicable in this case because their
acteristics, which might result in miss-detection. In the fi-
values do not change at all. Temperature sensor, acoustic
nal step, the min-max detection algorithm results in a value
sensor, and the x-axis of the magnetometer also could not
1 for a passing car and 0 for no cars, thus detecting all the
catch a car. The best sensors that detect the car are the y-
cars successfully.
axis of the magnetometer, the ultrasonic sensor, and visual
light with laser pointers. In Table 3 we present the summary
5.2.2. Ultrasonic Sensor
of these sensor characteristics. We also tested for hardware
Two ultrasonic sensors, SRF02, are mounted on Tmotes and
variations of the same sensing modality. In particular, five
attached to the ceiling on the fourth floor of the parking lot
magnetometers are tested in the same environment and the
to detect cars going up and down, as shown in Figure 9.
variance between four of them is found to be very large. In
Sensor 1 detects cars going up and sensor 2 detects cars
addition, one of them shows very little change in the mag-
going down. Because an ultrasonic sensor returns the dis-
netic field when a car passes.
tance of a sensed object from itself, the returned distance is
shorter when a car passes. It is even possible to distinguish
5.2. Detection Experiments vehicles from persons. Sensing with same frequency (70
5.2.1. Magnetometer ms sensing period), a person is detected for much shorter
Based on the y-axis measurements of the magnetometer, duration that a car, although a car is faster than a person.
the modified min-max algorithm is implemented to detect Figure 10 presents the results of our experiment. Note that

52

Authorized licensed use limited to: University of Southern California. Downloaded on November 25, 2009 at 15:56 from IEEE Xplore. Restrictions apply.
 (a)

(b)

(c)

(d)

(e)

Figure 8. Results at Each Step of the Modified Min-max Algorithm

53

Authorized licensed use limited to: University of Southern California. Downloaded on November 25, 2009 at 15:56 from IEEE Xplore. Restrictions apply.
 Table 3. Summary of Sensor Characteristics

Sensor Type Characteristics

- seriously influenced by environmental noise
Acoustic
- detection algorithm is difficult
- too sensitive, always maximum at day time
Visual Light
- cannot be used at night because of ceiling lights
Infrared - too sensitive, always maximum at day time
Temperature - no change in measurements, hard to detect vehicles
- reacts to only certain objects (cars, not humans)
Magnetometer - location dependent, raw data differs quite a lot based on locations
- relatively difficult detection algorithm
Ultrasonic - detection is easy and reliable
- detection is easy
- difficult to distinguish cars and humans
Visual light with infrared
- could be used as secondary sensors
- laser is always turned on, thus consuming lot of energy

the graph even shows the shape of the car. 5.4. Car Counting Experiments
Finally, we conducted experiments to count the number of
cars moving in and out at the one of the two entrances of
5.3. Transmission Experiments the university parking lot over a day. Ultrasonic sensors are
installed on the ceiling of the entrance. When a passing
We conducted several experiments inside the parking lot to
car is detected, the motes connected with sensors record the
get estimates about the link qualities between the sensors
time and passing direction into its flash memory. The results
on different floors. A good link quality is essential for the
during one day of the experiment are shown in Figure 12.
sensors to efficiently collaborate their individual measure-
As evident, the number of cars increases from morning 8am
ments and send the consolidated results to a base station.
until 2pm, and then decreases until 11pm. There are rarely
In particular, we measured the packet reception rates (PRR)
moving cars during the night time from 11pm to 6am except
and the received signal strength (RSSI) between two nodes
for periodic patrol cars.
transmitting at maximum power, as shown in Table 4. Our
results, presented in Figure 11 show that the nodes cannot
communicate with each other unless they are on the same 6. CONCLUSIONS
or adjacent floors. Transmission is seriously hindered by
obstacles, such as parked cars and walls, thus, causing rout- In this paper, we have argued the use of both ultrasonic and
ing problems. To overcome this, additional motes, possibly magnetic sensors in accurate and reliable detection of ve-
without sensors that could act as intermediate forwarders, hicles in parking lots. We have also proposed detection al-
are needed. gorithms for both these sensing modalities. Our extensive
experimental results show that the most effective of the sen-
sors are the magnetometers in the SBT80 sensor board, and
Table 4. PRR and RSSI Measurement Results
the ultrasonic sensor, SRF02. In addition, acoustic sensors
Link PRR RSSI could be used as secondary sensors to provide specific char-
acteristics of the vehicle and help a primary sensing method.
A - D (along the roadway) 0.67 -88.35
Each kind of vehicle has its own characteristics, such as
A-B 0.72 -85.64
body shape, magnetic wave pattern, engine sound pattern,
D-E 0.54 -85.01
etc. One could create a database of these characteristics
A-C 0 none
for each vehicle and make the detection algorithm match
D-F 0 none
their patterns, thus, being able to detect and identify spe-
B-E 0 none cific types of vehicles. It is also possible to implement more
C-F 0 none energy efficient sensing techniques by changing the sensing
period dynamically. For instance, in the normal mode an
ultrasonic sensor could have a relatively long sensing pe-

54

Authorized licensed use limited to: University of Southern California. Downloaded on November 25, 2009 at 15:56 from IEEE Xplore. Restrictions apply.
 (a)

(b)

Figure 10. Results of Ultrasonic Sensor Experiments

Figure 12. Results of Car Counting Experiments

55

Authorized licensed use limited to: University of Southern California. Downloaded on November 25, 2009 at 15:56 from IEEE Xplore. Restrictions apply.
 (a)

(a)

(b) (b)

(c)
(c)

Figure 7. Raw Experimental Data from Magnetic(y), Ul-

trasonic, and Visual Light with Laser

(d)

(e)

Figure 6. Raw Experimental Data from Magnetic(x),

Acoustic, Temperature, Light, and Infrared Sensors
Figure 9. Ultrasonic Sensors Deployed on Ceilings

56

Authorized licensed use limited to: University of Southern California. Downloaded on November 25, 2009 at 15:56 from IEEE Xplore. Restrictions apply.
 [7] Cheung, S.Y., S.C. Ergen, and P. Varaiya, “Traffic Surveil-
lance with Wireless Magnetic Sensors,” Proceedings of the
12th World Congress on Intelligent Transport Systems, San
Francisco, CA, Nov 2005.

[8] Crossbow Technologies, http://www.xbow.com

[9] Devantech SRF02,

http://www.acroname.com/robotics/parts/R287-
SRF02.html

[10] Ding, J., S.Y. Cheung, C.W. Tan, and P. Varaiya, “Vehicle
Figure 11. Transmission Experiments
Detection by Sensor Network Nodes,” California Partners
for Advanced Transit and Highways (PATH), Research
riod; however, once it detects that the returned distance is Reports: Paper UCB-ITS-PRR-2004-39.
smaller than a pre-specified threshold, it reduces its sensing http://repositories.cdlib.org/its/path/reports/UCB-ITS-PRR-
period. After the object passes and the returned distance be- 2004-39
comes more than a threshold the sensor recovers its normal [11] EasySen SBT80, http://www.easysen.com/SBT80.htm
sensing period.
[12] IEEE 802.15.4 WPAN,
http://www.ieee802.org/15/pub/TG4.html
ACKNOWLEDGEMENTS
[13] IrisNet: Internet-scale Resource-Intensive Sensor Network
We would like to thank Professor Bhaskar Krishnamachari Service, http://www.intel-iris.net.
of the Autonomous Networks Research Group at USC, and
the anonymous referees for their valuable inputs and feed- [14] Knaian, A.N., “A Wireless Sensor Network for Smart
Roadbeds and Intelligent Transportation Systems,” Master’s
back, which have greatly helped to improve the quality of
Thesis, MIT, Jun 2000.
the paper.
[15] MIT Intelligent Transportation Systems, http://mit.edu/its
REFERENCES [16] Moteiv (Sentilla), http://www.sentilla.com
[1] Akyildiz, I.F., W. Su, Y. Sankarasubramaniam, and E. [17] O’Flynn, B., S. Bellis, K. Mahmood, M. Morris, G. Duffy,
Cayirci, “Wireless Sensor Networks: a Survey,” Computer K. Delaney, and C. O’Mathuna, “A 3-D Miniaturised Pro-
Networks, Elsevier, Vol. 38, No. 2, 2002, pp. 393–422. grammable Transceiver,” Microelectronics International,
Emerald Group, Vol. 22, No. 2, 2005, pp. 8–12.
[2] Akyildiz, I.F., D. Pompili, and T. Melodia, “Underwater
Acoustic Sensor Networks: Research Challenges,” Journal [18] Responsive Roadways,
of Ad Hoc Networks, Elsevier, Vol. 3, No. 3, 2005, pp. 257– http://www.media.mit.edu/resenv/vehicles.html
279.
[19] Siemens Sipark PMA, http://www.siemens.com/traffic,
[3] Barton, J., J. Buckley, B. O’Flynn, S.C. O’Mathuna, J.P. http://www.siemens.fi/CMSTSWWW.nsf/702C2575B19B4B
Benson, T. O’Donovan, U. Roedig, and C. Sreenan, “The 1DC225700D00486438/$file/Sipark GB.pdf
D-Systems Project - Wireless Sensor Networks for Car-
Park Management,” Proceedings of the 65th IEEE Vehic- [20] Tang, V.W.S., Y. Zheng, and J. Cao, “An Intelligent Car
ular Technology Conference, Dublin, Ireland, Apr 2007, pp. Park Management System based on Wireless Sensor Net-
170-173. works,” Proceedings of the 1st International Symposium on
Pervasive Computing and Applications, Urumchi, Xinjiang,
[4] Benson, J.P., T. O’Donovan, P. O’Sullivan, U. Roedig, C. China, Aug 2006, pp. 65–70.
Sreenan, J. Barton, A. Murphy, and B. O’Flynn, “Car-Park
Management using Wireless Sensor Networks,” Proceedings [21] TinyOS 2.x, http://www.tinyos.net
of the 31st IEEE Conference on Local Computer Networks,
Tampa, FL, Nov 2006, pp. 588–595. [22] Tyndall National Institute, http://www.tyndall.ie

[5] Boda, V.K., A. Nasipuri, and I. Howitt, “Design Considera- [23] Xu, N., S. Rangwala, K.K. Chintalapudi, D. Ganesan, A.
tions for a Wireless Sensor Network for Locating Parking Broad, R. Govindan, and Deborah Estrin, “A wireless sensor
Spaces,” Proceedings of IEEE SoutheastCon, Richmond, network For structural monitoring,” Proceedings of the 2nd
VA, Mar 2007, pp. 698–703. International Conference on Embedded Networked Sensor
Systems, Baltimore, MD, Nov 2004, pp. 13-24.
[6] Car-Park Management using Wireless Sensor Networks,
http://www.cs.ucc.ie/misl/research/carparking.php

57

Authorized licensed use limited to: University of Southern California. Downloaded on November 25, 2009 at 15:56 from IEEE Xplore. Restrictions apply.