Visualizing Sensor Data

Christian Richter

Abstract Visualizing sensor data is not a trivial task. This work tries to find a matching between sensor data and visualizations.
Sensors became very popular over the last decade, they got smaller, more efficient and cheaper, that is why they spread in many
fields of applications. Sensor taxonomy and the dependency of space and time will help to find visualizations for sensor data. Besides
this there are problems with visualizations that will be worked out and with the results of researchers it is possible to find a mapping
between sensors and visualizations. This mapping is discussed in the end of the work. It is a mapping from sensor data dimension
and the dependency of the spatio-temporal aspect to the dimension of the data of the resulting visualization. With this resulting
dimension it is possible (according to Shneiderman [17]) to create a fitting visualization.
Index TermsSensor, Visualization, Taxonomy, Data Fusion, Spatio-Temporal, Multidimensionality

1 I NTRODUCTION
The field of applications where sensors are used is spreading extremely is used. Finally in section 4 there will be an attempt in making
fast. The variety of sensors is increasing as much as their ease of use a matching between sensor data and visualizations. Previously
[5]. Since sensor data appears in large amounts and because of the challenges and problems will be discussed followed by possible
multidimensionality of the data it is important to analyze this data with starting points. Before the final matching there will be a short detour
technical help [5]. The next step after the analysis of the sensor data on multivariate and multiview visualizations as a reasonable solution
is to find a visualization for the data, which is very task specific. Vi- to visualize multidimensional data.
sualizations are important when working with sensor data, it makes it
more comfortable for a user to work with the data and the data can
be understood faster and easier. With a visualization it is possible
to find patterns, connections or similarities in numerical data. That
makes it a lot easier then to manually analyze the raw sensor data,
which is sometimes impossible to understand for a person. In Fig-
ure 1 there can be seen an example for a visualization. This one is
selected from the CarTel software. The CarTel system is using sen-
sors on mobile units to make analysis about the traffic data in certain
areas [11]. A possible visualization which can be extracted from the
sensed data is the presentation of traffic hot spots. On the right side
there is a map with marked positions and on the left side there is a
table showing the sensor values, which are linked to the marked po-
sitions on the map (see Figure 1). The visualization from CarTel is
divided in two parts (the map and the table). Now the arising question
is, whether it is possible to combine these two parts to a single visu-
alization, which makes it easier and faster to receive the information.
Sensor data has to be visualized, but as can be seen in CarTel System
that is not a trivial task. With this idea of visualizing sensor data, there Fig. 1. The CarTel portal, showing users traffic hot spots for the Seattle
are some upcoming questions: area [11]

How can sensor data be presented in a reasonable visualization?

2 S ENSORS
Is a mapping between sensor data and visualizations possible?
This section examines the current state of the art of sensors, sensor
Is it reasonable to map certain sensor kinds to a specific visual- data, sensor data fusion and the common ground of the spatio-temporal
ization? aspect in the sensed data.

Goal of this paper is trying to give guidelines for 2.1 Sensor Taxonomy
finding a fitting visualization for specific sensor data. Sensors are not only becoming popular in scientific tasks, they are
Section 2 will discuss the taxonomy of sensors and the unique spreading in every day life tasks. Since this development is pro-
properties of sensor data. Besides this the necessity of the data ceeding pretty fast, there are a lot of sensors nowadays [4]. In the
fusion process (aggregation and extraction) and the importance of the following a classification for the large number of sensors existing is
spatio-temporal component of sensor data will be another important shown and the fields where this sensors are used are presented. Using
point. Section 3 will have a closer look at visualizations, how those Whites [20] classification scheme results in the following two tables.
can be classified, what is important when designing a visualization One for the classification of the sensors and the other for the field of
and the important role of the user and the technical devices that applications where sensors are used (see Table 4 (at the end of the
paper) and Table 1). White is sorting the types of sensors by their
measurand. This classification can help a programmer when imple-
Christian Richter is studying Media Informatics at the University of menting a visualization. With knowledge about the sensed value he
Munich, Germany, E-mail: christian.hans.richter@campus.lmu.de can use fitting color scales, create specific shapes and so on (com-
This research paper was written for the Media Informatics Advanced pare section 3.2). The table with the field of applications can help
Seminar on Information Visualization, 2008/2009 to improve a mapping between sensor data and visualization, since
it is possible to resort to previous visualizations used in similar task.
As can be seen in the table, sensors are deployed in a lot of fields,
 2.3 Data Fusion
Table 1. Field of applications for sensors [20]
In sensor networks large data sets have to be handled. This
Field of Applications can be put down to the enormous number of sensor nodes
Agriculture which can be used and to the continuous sensor values [1].
Automotive Data Fusion is a task that arises from this large data sets.
Civil engineering, construction With such large data sets there has to be a quality check of the
Distribution, commerce, finance data before making further proceedings. Sometimes it can happen
Domestic appliances that sensors are not correctly calibrated or that a sensor malfunc-
Energy, power tions. These values have to be cleaned or excluded in further pro-
Environment, meteorology, security ceedings [5]. After the collection of the complete data set (aggre-
Health, medicine gation) and the removal of incorrect values, there has to be an ex-
Information, telecommunication traction of the data of interest. That means an extraction of certain
Manufacturing sensor values[19]. This two processes the aggregation and extrac-
Marine tion of sensor data will help to find patterns in large sets of data.
Military
Having a great number of temperature sensors delivering continu-
Scientific measurement
ous sensor values in a certain terrain and a certain time, it can be
Space
Transportation (excluding automotive)
interesting for a user whether or not a certain threshold is reached.
Other (specify) Since the user only needs a visualization of this threshold and fur-
ther information like when and where it happened, the total data
set of all sensors has to be aggregated (for example in a database)
and the values of interest have to be extracted. If there are any
values of interest then this will be visualized in an application.
which gives a lot of visualizations already existing. Whites taxon- Another task for data aggregation from multiple sensors is radar or ul-
omy was made in 1987, what means that it was created about 20 trasound examination. In these tasks there are multiple sensors used.
years ago, so it is rather old. To consolidate his taxonomy it can Their data has to be aggregated before visualizing it, since the data of
be compared to [12], which is saying nearly the same as was said a single node would not make much sense to be displayed. [19] says
in 1987. There havent been any great changes and developments in that the data has to undergo some preprocessings, these are:
the classifying, neither on the sensors nor on the field of applications.
This should make it a reasonable taxonomy for sensors nowadays. Feature extraction
Along with dividing the sensors by its measurand, they can also be
watch by functionality. On the one hand, there are passive sensors. Data cleaning
These are able to generate electric signals without an additional elec-
trical source. On the other hand, there are active sensors. To fulfill Data reduction
their task they need an external power source. Detailed and additional
information on this way of classification can be found in [7]. Dimension reduction

Feature extraction will select the relevant attributes or in other

words the data of interest. Data cleaning will improve the qual-
2.2 Sensor Data
ity of the data by reducing noise or excluding errors. Data re-
A sensor is a device that receives a stimulus and responds with duction represents the data aggregation process. Dimension re-
an electrical signal[7]. The stimulus or measurand, was al- duction is obviously the reduction of the number of features.
ready observed in the previous section. These are usually phys- All in all data fusion is important for a user, once more expressed by
ical or chemical forces or processes. In this part the electrical the following citation: Users will routinely require compressed sum-
signals, with which a sensors responds, are watched. The Hand- maries of large spatio-temporal sensor data. However, periodically or
book of Modern Sensors [7] says, that sensors measure (in most occasionally, users will require detailed datasets from a subset of sen-
tasks) non-electronic values and transfer these into electrical val- sors in the network[8]. This will lead to another important fact about
ues. According to section 2.1 these values can be sensed in an sensor data, the spatio-temporal aspect.
active or passive way. Since in this section the data of a sen-
sor is the interesting thing, it can be classified in another way. 2.4 Spatio-Temporal Component
Measuring a value with an absolute scale, an absolute sensor has Large sensor networks often have a lot of sensor nodes. When
to be used to fulfill this task. A possible example would be the watching such networks it is important to know where and when
sensing process of a temperature value according to the Kelvin these nodes measured their values. This is necessary because the
temperature scale, which is totally independent of the measur- sensed value has to be connected to a certain point of the net-
ands. Beyond the absolute sensors there are the relative ones. work and must have a time value. Without having this informa-
A relative sensor generates its data according to a referencing tion it is not possible to make any predictions or conclusions [14].
value. Imagining to sense pressure a relative sensor can be This will become clearer when watching an example. There is a
used which senses the pressure in relation to the air pressure [7]. temperature sensor which measured a temperature value of 40 degree
That means, that an application interacting with sensors needs to know Celsius. Without additional information it is only known that some-
with which kind of sensor it interacts, whether it is active or passive. where at some time a temperature of 40 degree Celsius was measured.
Also it needs to know if it sends continuous data or if the data is deliv- Putting the sensed value in the context of a desert, then it wouldnt
ered on request [1]. Wu says: However, different sensors may use dif- surprise too much that such a value was measured. Putting it in the
ferent physical principles, cover different information space, generate context of the north pole it would be an extraordinary value. So the
data in different formats at different updating rates, and the sensor- location where a sensor is used is of great importance. Staying at the
generated information may have different resolution, accuracy, and example of the temperature sensor and assuming that the sensor is used
reliability properties.[21]. That is to say, depending on the task and in a desert, then time plays an important role, too, because tempera-
the used sensors, different types of data will occur and the data will ture values can vary a lot during day- and nighttime in deserts [16].
have to be computed in different ways. All in all the sensor data is Having a closer look at the importance of the position of a sensor it
only useful for a user, if he knows the context where the data is sensed is necessary to say, that absolute position must be distinguished from
and if he knows which data the sensor conveys. relative position. Large sensor networks have multiple nodes which
could be spread over a certain area. In this case the absolute posi- presenting data in visualizations are numerous. According to the di-
tion of a sensor is relevant because it would not help much to say, mensions of the data the dimensions of the visualization has to be ex-
the sensor is at the hillside, it is necessary to know its coordinates. tended. That means, that if there is a 1-dimensional set of data, then
When watching more then one sensor node the absolute position can the dimension of the visualization does not need to be more then 1
be important when wanting to calculate the speed of an object moving (but it can be). All in all it is to say, that there exist a large number of
between these two nodes. Besides the absolute position there is the rel- visualization techniques and that they develop every day so a listing of
ative one. Relative sensors are attached to objects or persons. The in- all would be impossible. That is why the taxonomy was made by the
teresting fact is at which object the sensor is, not its absolute position. dimensions.
A possible example for a relative sensor would be a door sensor [16].
As already said, time has an important role, too. This has multiple rea- 3.2 Visualization Guidelines
sons. First, it is important to have knowledge about the time when a This section examines two points. First, there are certain prin-
sensor measured a value to make better statements about the data. Sec- ciples that can help in designing visualizations, such as shape,
ond, time is a synchronization tool in sensor networks, therefore it is color, etc. which apply to human cognition. Second, there
relevant to have a clock within a sensor. Lastly when watching a sen- are some representative scales in every day life for visualiz-
sor, it can be used to show momentary data or to recognize continuous ing special things like temperature, which is normally displayed
data over a certain time period [16]. with blue for cool temperatures and with red for hot ones.
According to [10] there are certain perception properties every per-
3 V ISUALIZATIONS son has in common. For example if it is about finding an object in a
Analysis of data and getting an overview of large data sets are tasks large amount of objects. If the object has a unique color or a unique
that have to be done in every day life. Data like sensor data normally shape it can be found pretty fast. These perception rules should be
exists in numerical values so the process of understanding or analyzing kept in mind when trying to design a user interface of good quality.
it is not trivial, finding patterns seems hardly possible. To make this Using them will make it easier for a user to find outliers in a large data
tasks easier there are visualizations. Visualizations take the numerical set. This and similar tasks can be simplified with the utilization of per-
data, analyze it, use extraction and aggregation methods and then give ception rules. A great overview of a lot of rules is given in [10], since
the user a graphical representation of the input data. This representa- the listing and discussion of all of them would go beyond this work.
tion makes it easier to understand the data and to interact in the data Apart from that, some typical design patterns occur in every day
set. In this paper the visualization limit to sensor data. The following life. These patterns are not forced to be used but in most cases
sections will discuss a visualization taxonomy and starting points for they are used since it is common sense. The most popular exam-
visualizations. ple will possibly be the mapping of temperature to a color scale
from blue to red. Another example would be brown and green
3.1 Visualization Taxonomy scales which represent height values on maps. So when design-
Making a taxonomy for visualization is a difficult task. Since there ing a visualization for sensors which are measuring temperature val-
are a a lot of visualization methods to present the same information ues it would be reasonable to use the mentioned color scale. An-
it is not trivial to find a taxonomy for visualizations. In this work other example would be a traffic light, where the color red rep-
there will be a classification-scheme according to Shneiderman, who resents the signal for stopping and green is the signal for going.
has done research work on the topic of information visualization. [17] Combining the principles of human cognition with the common ways
says that there are seven different types of data to be visualized. That of visualizing certain data will help to create good and easy to under-
is a possible way to classify the types of visualizations, according to stand visualizations.
its underlying dataset. The seven data types are:
3.3 Customized Visualizations
1-dimensional As already shown in the previous section there are a lot
of things that have to be regarded when visualizing informa-
2-dimensional tion. In this section the user and the hardware will be dis-
cussed in short as another starting point to create visualizations.
3-dimensional One great factor for the design of a user interface is the interface,
in other words, the hardware. Showing a map on a PDA is not
Temporal
easy since the PDA has a rather small display and the information
Multi-dimensional shown on it has to be reduced to a certain amount so that it is easy
to understand. Showing the same map on a large (television-like)
Tree display there can be shown much more information as will be dis-
played on a PDA. The field of applications for sensor data is wide
Network spread (compare section 2.1) so it is likely that the hardware varies.
Watching the CarTel example from the beginning [11] the data that
1-dimensional data types are every form of linear data for example is sensed can be displayed on a normal device like a PC or a note-
text documents. The main principle is that the data is organized in a book. Taking a closer look, the data gathered from the sensors is traf-
linear way. 2-dimensional is for example map-data like geographical fic data and is interesting for the participants of the traffic system. Be-
maps or something similar. Data in this type have planar character cause of this it makes sense to display the data in an in-car-system.
and can be projected on a map. 3-dimensional refers to real objects, This is only one example for the importance of customized visualiza-
Shneiderman says, that its about objects with volumes with some tions depending on the hardware used for showing the information.
potentially complex relationship to other items [17]. Temporal: This Besides the hardware there is the user himself who needs to be re-
data has a start and finish time and the objects can be projected to a garded for the development of a visualization. A user can be un-
timeline. Multi-dimensional data is mostly data in databases, as sensor educated in the topic which will be visualized. That is the rea-
data is stored in databases, too, and mostly has a lot of attributes this son why it is important, that the visualization consists of low com-
can be of interest in further observations. Tree: The underlying data plex parts, that there is additional information on demand, and
has a hierarchical structure and can be display in the form of a tree. further methods to add information. Having a specialist watch-
Finally there are networks. They are similar to trees, that means there ing the same data, a different visualization could make much
is a connection between the data elements, but not in a hierarchical more sense, since he has much more knowledge about the topic.
way. Possible visualizations for each data type are presented in [17]. Interface and user are one big factor that influences the creation of
Like in Readings In Information Visualization [18] said, the ways of a visualization. Additionally there are the guidelines of the previous
section which have to be fulfilled. With this two big problems the good visualization. There are four steps, that have to be done, when
design of a visualization is not a trivial task. creating a design, these are:

4 M ATCHING BETWEEN S ENSOR DATA AND V ISUALIZATION Overview first,

The final part of this work is to find a possible matching between zoom and filter then
sensors and visualizations. Before doing this, there has to be a short
overview of problems which can occur when trying to visualize sensor details-on-demand
data. Trying to connect the topics sensor data and visualization to give
These four steps are explained in detail in [17]. The main princi-
advice for the visualization of sensor data is the goal in this section.
ple is to give an overview of the whole data set. To take a closer
4.1 Challenges and Problems look at special data there a tools that allow zooming functions or
which filter away uninteresting data. If the user wants to have de-
Trying to visualize sensor data there will be a confrontation with chal- tails about the data, there should be a possibility to get these details.
lenges and problems. A closer look at sensor networks will be made To get a better understanding of this four steps there will be a
to explain this in more detail. [4] is saying that when having a large short example. A sensor is measuring a lot of values, for exam-
network of sensor nodes and there are for example two targets which ple temperature, barometric pressure, humidity and so on. Then
are watched by the sensors, then each node in the network is giving all these values are displayed in one big graph (that would be
a different identifier to one of the objects. When trying to visualize the overview). Zooming in on a special week and filtering out
the data later a programmer needs to know which data is referring to all values but temperature would refer to zoom and filter. Lastly
which object. Another challenge is the sensor network itself. The life- if the user is interested in a single value on a special day of
time of such a network is limited by the energy resources of the nodes. this week, this would be the details on demand, where addition-
When there is a lot of communication or processing then the nodes ally the other values can be shown, too. Shneidermans Mantra
will be unable to sense data for long periods. Along with the lack of was a second starting point to find or create a good visualization.
energy there is the problem of the reliability. That means the prob- A third point would be the spatio-temporal aspect of sensor data (sec-
lem of noise in sensor data or the collecting of wrong values. All this tion 2.4). In short words, there are special ways to visualize spatial
problems occur when trying to visualize the sensor network data [4]. data and temporal data[18]. This visualization techniques can be used
According to [5] there are additionally some unique proper- as a starting point to create an own visualization.
ties of sensor data which will make problems in the pro-
cess of visualizing them, partly they are covering with [4].
The two largest problems of sensor data are, that they appear in
a large amount and that the data itself is multidimensional. An-
other problem is, as already mentioned, noise or failure of sen-
sors. Since it can happen that sensors stream their data, this
has also to be taken in regard when implementing a visualiza-
tion. [5] and [4] both say, that sensor data can sometimes be
gathered on-the-fly according to an example of Chong this can
happen when an airplane flies over a surveillance area and sends
a query to the sensor nodes. It can not been taken for granted
that there is a 100% reliability of the sensors to answer the query.
Beyond the sensors itself there are problems when design- Fig. 2. Avalanche rescues: Screen sketch for liquid crystal displays [13].
ing visualizations. The biggest problem according to [15] is
the fitting of a visualization to the wishes of the user, to
the task and to real world problems. Trying to display the
data there should be different views on the same data set.
This features have to be implemented and the properties of sensor data
have to be regarded to have a good visualization of sensor data. As can
be seen there are a lot challenges when visualizing sensor data.

4.2 Starting Points

There are multiple starting points when trying to find a visualization
for sensor data. This work will take a look at some of them, partly
described in earlier sections. One possibility to start would be the
taxonomy of the sensors watching the sensor task in special. This
will become clearer with the following examples. The first one is
about wearable sensors which should help within avalanche rescues
[13]. As can be seen here, the interesting thing is a single sensor
value which is showing the position and the vital signs of the vic-
tim buried in snow. So in this case a visualization will need a single
sensor to be displayed. A possible visualization was given in the pa-
per (compare Figure 2). The second example is about understanding
the environment, therefore there are multiple sensors used [6]. If this
data will be displayed the task is different from the first one because
the interesting data is not located in a single node. It is distributed Fig. 3. Overview over three sensor pods(each color represents a pod)
in a large number of nodes. To make sense of the total data, there [6]. Graphs show (top to bottom): surface temperature (degree Celsius),
has to be shown more then the values of one node (as can be seen in surface moisture, and soil moisture at 0.5 m depth
Figure 3). As can be seen, there is a dependency between the task
and the visualization, so the task of the sensor is one starting point.
Another starting point can be found in the topic of information visu- 4.3 Multivariate and Multiview Visualization
alization. According to Shneiderman there is the Visual Information Multidimensionality is one of the biggest problems when visual-
Seeking Mantra [17]. He says, that this is a starting point to create a izing sensor data [5]. A lot of sensors measure more then one
 is representing the spatio-temporal dependency of the sensor value.
Table 2. High-dimensional visualizations[9]
The dimension of the measurands serves as a base for the resulting
data type (compare to section 3.1). If the dimension of the mea-
Visualization type
surand is 1 then the data type is 1-dimensional, too, according to
2D and 3D scatterplots
Shneiderman [18]. Adding the spatio-temporal aspect of sensor data
Matrix of scatterplots
Heat maps
to the dimension of the measurand, the final data type will be ex-
Height maps tended. Referring to section 2.4 the position of a sensor is abso-
Table lens lute or relative. Having a relative position, it is not necessary to ex-
Survey plots tend the dimension of the data type. A simple label with informa-
Iconographic displays tion about the relative position is enough. Absolute position how-
Dimensional stacking (general logic diagrams) ever extends the dimension. Assuming to have 2-dimensional posi-
Parallel coordinates tion values, it is necessary to display the x- and y-position of the sen-
Line graph, multiple line graph sor data value. So, the data type dimension will be extended by 2.
Pixel techniques, circle segments Besides the position there is time as another influence factor on sen-
Multi-dimensional scaling and Sammon plots sor data. Also shown in section 2.4 sensors measure values con-
Polar charts tinuous or momentary. A sensor, which registers momentary val-
RadViz ues does not extend the dimension of the resulting data type, since
PolyViz the information is always displayed immediately on the screen and
Principal component and principal curve analysis no temporal information has to be encoded. Continuous sensors do
Grand Tours have to record information over time and these values have to be dis-
Projection pursuit played. A timeline will extend the resulting data type dimension by 1.
Kohonen self-organizing maps The measurands dimension together with the spatio-temporal aspect
can be mapped to Shneidermans taxonomy by data type. This tax-
onomy gives advice for a possible visualization [17]. The resulting
matrix (compare table 3) contains four data types 1-dimensional, 2-
kind of data at a time. For example it senses temperature, air dimensional, 3-dimensional and multi-dimensional (abbreviated by
pressure, humidity and so on. Displaying all these values of 1-d, 2-d, 3-d and multi-d). For each of this data types Shneider-
the sensors from a large area is getting really difficult. This man is presenting ideas and possibilities how to visualize the data.
leads to the usage of multivariate and multiview visualizations. An example from the earlier presented avalanche rescues is showing
The official definition of a multiple view system, according to Bal- a sensor which is measuring the heart rate among other values. The
donado, is: We say that views are distinct if they allow the user to sensor data itself is 1-dimensional (only regarding the heart rate). Be-
learn about different aspects of the conceptual entity, e.g., by pre- sides the sensor data there is the position of the sensor which is relative
senting different information, or by emphasizing different aspects of in this case (it is only interesting who is wearing the sensor) and the
the same information. A multiple view system uses two or more temporal aspect of the sensor value which is continuous. According
such distinct views to support the investigation of a given concep- to the matrix the resulting data type is 2-dimensional since the input
tual entity[3]. Returning to the weather example, that would mean, values are 1-dimensional sensor data and relative-continuous depen-
that there is a view showing the temperature data of a single sensor. dency. With the proposal for a data type it should be possible to create
Another view displays the sensors humidity value and a third view a visualization (compare [17]). Figure 4 is showing a visualization
could for example show a map of all sensors and their current tem- chosen in [13] to display the heart rate over a certain time.
perature values. The target of this multiple view displays is to re-
duce the dimension of the data by showing it in more then one view.
Multivariate visualizations are another way to present n-dimensional
data sets besides the multiple view displays [9]. That makes
them important for the presentation of sensor data, which are of-
ten, like seen earlier in this paper, of multidimensional charac-
ter. Table 2 is adopted from [9] and gives an overview of high-
dimensional visualizations which mostly can be used for the vi-
sualization of sensor data. Grinstein is presenting detailed in-
formation on all named visualizations in his paper [9]. Since
this work is not handling multivariate visualizations and the in-
formation is not relevant for the mapping it is not shown here.
The importance of multiview and multivariate visualizations is not
only given because of the multidimensionality of the data, but addi-
tionally because of the dependency of time and space of the measured
value. An absolute location value can have up to three dimensions
which are interesting for the user. Those will have to be presented in a
visualization plus the temporal aspect. That means, that the visualiza- Fig. 4. Oximeter measurements [13] (in this case only the heart rate is
tion techniques presented, in short, in this section are really important. of interest)

4.4 Mapping
This last section is trying to provide a mapping between sensor 5 C ONCLUSION
data and visualizations. This mapping is a new contribution and is Having found a possible mapping from sensor data to visual-
created from the results of the previews work, most influenced by izations in this work, it would be recommended to prove this
Shneidermans data type classification which was presented in sec- matching with a large study. Different combinations of sen-
tion 3.1. Table 3 is showing a possible mapping from sensor data sor data with different dependencies of space and time should be
to visualizations. Its idea is to project the dimension of the mea- tried out and be visualized according to the resulting data type.
surands and the dependency of space and time on the visualiza- Since Shneiderman only gave a taxonomy by data type for the vi-
tion taxonomy by data type from Shneiderman [17]. One axis of sualizations [17] the final word of how to implement the visualiza-
the table is representing the dimension of the measurands, the other tion is not yet spoken. Some of the mentioned starting points can
Table 3. Mapping of sensor data to visualizations. The cell entries represent the dimension of the resulting data type of the visualization according
to Shneiderman [17].

1-dimensional 2-dimensional 3-dimensional multi-dimensional

relative-momentary 1-d 2-d 3-d multi-d
relative-continuous 2-d 3-d multi-d multi-d
absolute-momentary 3-d multi-d multi-d multi-d
absolute-continuous multi-d multi-d multi-d multi-d

help in developing visualizations. But since the field of sensors grew

pretty fast over the last years (compare [4]) it is probable that in the Table 4. Sensor classification by measurands [20]
future sensors will be different from sensors nowadays. This could
lead to totally different ways of visualizations, which are not think-
able of right now. Besides the sensors there is the visualization hard- super category sub category
ware which will introduce new possibilities in future. This technical Acoustic
progress makes it difficult to give a mapping between this two topics. Wave amplitude, phase, polarization, spectrum
An up to date example of this changes is augmented reality. Aug- Wave velocity
mented Reality is briefly said the extension of the real world with Other (specify)
virtual elements [2]. To extend the real world there need to Biological
be sensors in the users environment that can sense his behavior Biomass (identities, concentrations, states)
so that an interaction between real world and computer is possi- Other (specify)
ble. The sensed behavior will undergo certain processes and fi- Chemical
nally there will be some output in different kinds of visualizations. Components (identities, concentrations, states)
Other (specify)
All in all it can be said, that hardware (both sensors and com-
Electrical
puters) and software is developing really fast and tasks extend to
Charge, current
a lot of fields (for example the augmented reality). Beyond this
Potential, potential difference
sensors are used in different tasks and in different ways, what Electric field (amplitude, phase, polarization, spectrum)
will make a unique visualization of a specific sensor impossible. Conductivity
As a conclusion there can be said that it is possible to make an assign- Permittivity
ment from sensors to visualizations regarding the dimension of the Other (specify)
sensor data and the data type of the visualization. However it is not Magnetic
possible to make statements about the look and feel or the interaction Magnetic field (amplitude, phase, polarization, spectrum)
methods of the visualization. It is advisable to design a visualization Magnetic flux
that resorts to typical every day life and widely used visualizations, Permeability
which an ordinary user can easily understand. Other (specify)
Mechanical
Position (linear, angular)
Velocity
R EFERENCES Acceleration
[1] I. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. Wireless Force
sensor networks: a survey. Computer Networks, 38(4):393422, 2002. Stress, pressure
[2] R. Azuma et al. A Survey of Augmented Reality. PRESENCE- Strain
CAMBRIDGE MASSACHUSETTS-, 6:355385, 1997. Mass, density
[3] M. Baldonado, A. Woodruff, and A. Kuchinsky. Guidelines for us- Moment, torque
ing multiple views in information visualization. In Proceedings of the Speed of flow, rate of mass transport
working conference on Advanced visual interfaces, pages 110119. ACM Shape, roughness, orientation
Press New York, NY, USA, 2000. Stiffness, compliance
[4] C. Chong and S. Kumar. Sensor networks: evolution, opportunities, and Viscosity
challenges. Proceedings of the IEEE, 91(8):12471256, 2003. Crystallinity, structural integrity
[5] D. Cook. Making Sense of Sensor Data. IEEE PERVASIVE COMPUT- Other (specify)
ING, pages 105108, 2007. Optical
[6] K. Delin, S. Jackson, D. Johnson, S. Burleigh, R. Woodrow, M. McAuley, Wave amplitude, phase, polarization, spectrum
J. Britton, J. Dohm, T. Ferre, F. Ip, et al. Sensor Web for Spatio-Temporal Wave velocity
Monitoring of a Hydrological Environment. In 35th Lunar and Planetary Other (specify)
Science Conference, March 15-19, 2004, League City, Texas, abstract, Radiation
2004. Type
[7] J. Fraden. Handbook of Modern Sensors: Physics, Designs, and Appli- Energy
cations. Springer, 2004. Intensity
[8] D. Ganesan, D. Estrin, and J. Heidemann. DIMENSIONS: Why do we Other (specify)
need a new Data Handling architecture for Sensor Networks? Thermal
[9] G. Grinstein, M. Trutschl, and U. Cvek. High-dimensional visualizations. Temperature
In Proceedings of the 7th Data Mining Conference-KDD, 2001. Flux
[10] C. G. Healey. Perception in Visualization. 2007. Specific heat
[11] B. Hull, V. Bychkovsky, Y. Zhang, K. Chen, M. Goraczko, A. Miu, Thermal conductivity
E. Shih, H. Balakrishnan, and S. Madden. CarTel: a distributed mobile Other (specify)
sensor computing system. In Proceedings of the 4th international con- Other (specify)
ference on Embedded networked sensor systems, pages 125138. ACM
Press New York, NY, USA, 2006.
[12] R. Luo, C. Yih, and K. Su. Multisensor fusion and integration: ap-
proaches, applications, andfuture research directions. Sensors Journal,
IEEE, 2(2):107119, 2002.
[13] F. Michahelles, P. Matter, A. Schmidt, and B. Schiele. Applying wear-
able sensors to avalanche rescue. Computers & Graphics, 27(6):839847,
2003.
[14] D. Niculescu et al. Positioning in ad hoc sensor networks. IEEE Network,
18(4):2429, 2004.
[15] C. Plaisant. The challenge of information visualization evaluation. In
Proceedings of the working conference on Advanced visual interfaces,
pages 109116. ACM New York, NY, USA, 2004.
[16] K. Romer and F. Mattern. Towards a unified view on space and time in
sensor networks. Computer Communications, 28(13):14841497, 2005.
[17] B. Shneiderman. The eyes have it: a task by data type taxonomy for in-
formationvisualizations. In Visual Languages, 1996. Proceedings., IEEE
Symposium on, pages 336343, 1996.
[18] B. Shneiderman, S. Card, J. Mackinlay, and B. Shneiderman. Readings
in Information Visualization: Using Vision to Think. Morgan Kaufmann,
1999.
[19] P. Tan. Knowledge Discovery from Sensor Data. SENSORS-
PETERBOROUGH-, 23(3):14, 2006.
[20] R. White. A Sensor Classification Scheme. Ultrasonics, Ferroelectrics
and Frequency Control, IEEE Transactions on, 34(2):124126, 1987.
[21] H. Wu. Sensor Data Fusion for Context-Aware Computing Using
Dempster-Shafer Theory. PhD thesis, Carnegie Mellon University, 2003.