Scholarly article on topic 'Determining Truck Activity from Recorded Trajectory Data'

Determining Truck Activity from Recorded Trajectory Data Academic research paper on "Civil engineering"

CC BY-NC-ND
0
0
Share paper
OECD Field of science
Keywords
{"Transportation fleet management" / "Route monitoring" / "Spatiotemporal description languages"}

Abstract of research paper on Civil engineering, author of scientific article — Zoltán Fazekas, Péter Gáspár, Roland Kovács

Abstract It is a well-known and understandable intention of freight transportation companies worldwide to verify the proper use of their truck fleets. An important and a hardly avoidable step of monitoring truck routes is the visualization of collected route data. To improve the efficiency of visual monitoring, two preprocessing steps are proposed in the paper. The first of these is segmentation of routes into meaningful route sections; while the second is the automatic rendering of route sections to vehicle activity classes. The route segmentation method is based on spatiotemporal features of the trajectory. It is a linguistic approach that can cope with the multiple spatial and temporal resolutions necessary to characterize vehicle/driver activities. These activities range from fairly simple activities to complex maneuvers. Pointers to the detected activities and maneuvers in the trajectory data can be used for indexing purposes. The low-speed vehicle maneuvers and stoppages are looked at in particular. The rendering of route sections to vehicle activity classes is based primarily on the recorded trajectory and speed data, but transportation specific and general road data can be also taken into account. The transportation specific data may include geographic and descriptive data on the actual and regular truck destinations, as well as on roadside accommodations and refueling stations. The general road data provide designations of roads, maximum speed allowed and the socio-cultural structure (e.g., urban, rural) along the roads. The preprocessed truck trajectories could facilitate the detection of traffic rule infringements and suspect driver behavior.

Academic research paper on topic "Determining Truck Activity from Recorded Trajectory Data"

Available online at www.sciencedirect.com

*%4 ScienceDirect

Procedía Social and Behavioral Sciences 20 (2011) 796-805

14th EWGT & 26th MEC & 1st RH

Determining Truck Activity from Recorded Trajectory Data

Zoltân Fazekasa*, Péter Gäspära, Roland Koväcsb

aComputer and Automation Research Institute (MTA SZTAKI), Kende u. 13-17, Budapest, H-1111, Hungary bKnorr-Bremse Braking Systems Ltd., Szegedi ut 49, Kecskemét, H-6000, Hungary

Abstract

It is a well-known and understandable intention of freight transportation companies worldwide to verify the proper use of their truck fleets. An important and a hardly avoidable step of monitoring truck routes is the visualization of collected route data. To improve the efficiency of visual monitoring, two preprocessing steps are proposed in the paper. The first of these is segmentation of routes into meaningful route sections; while the second is the automatic rendering of route sections to vehicle activity classes. The route segmentation method is based on spatiotemporal features of the trajectory. It is a linguistic approach that can cope with the multiple spatial and temporal resolutions necessary to characterize vehicle/driver activities. These activities range from fairly simple activities to complex maneuvers. Pointers to the detected activities and maneuvers in the trajectory data can be used for indexing purposes. The low-speed vehicle maneuvers and stoppages are looked at in particular. The rendering of route sections to vehicle activity classes is based primarily on the recorded trajectory and speed data, but transportation specific and general road data can be also taken into account. The transportation specific data may include geographic and descriptive data on the actual and regular truck destinations, as well as on roadside accommodations and refueling stations. The general road data provide designations of roads, maximum speed allowed and the socio-cultural structure (e.g., urban, rural) along the roads. The preprocessed truck trajectories could facilitate the detection of traffic rule infringements and suspect driver behavior. © 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of the Organizing Committee.

Keywords: Transportation fleet management; Route monitoring; Spatiotemporal description languages.

1. Introduction

It is a well-known and understandable intention of freight transportation companies with truck fleets to verify the proper and lucrative use of trucks. An important and a hardly avoidable step of monitoring truck routes is the visualization of collected route data. Apart from other useful measurement data, route data normally refer to discrete position and speed measurements taken at regular intervals. For haulage operations of medium size and above, the human inspection of the recorded trajectory data is not a viable option as it is highly time-consuming and unreliable. Therefore, semi-automated tools, and/or fully automated route monitoring systems and services are required and in many cases used by these companies. Such systems and services can work either in an online, or offline manner. The concrete system should be chosen to meet the exact needs of the transportation company. In spite of the

* Corresponding author. E-mail address: zoltan.fazekas@sztaki.hu

1877-0428 © 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of the Organizing Committee doi:10.1016/j.sbspro.2011.08.088

available tools and systems, there is still place for improvements in respect of monitoring efficiency in case of companies with truck fleets of moderate size. Measurement data gathered from a small fleet of trucks running in North America over a period of few weeks served as test data for the methods and approaches presented herein and helped us in understanding the characteristic behaviors/routes of drivers/trucks. Working with the measurement data improved our understanding the needs of the freight companies and helped explore the visualization possibilities.

Though the trajectory and speed data acquired from the Global Positioning System (GPS) is the only measurement data used herein - this data will be looked at and analyzed in a shipment and vehicle security context -a richer set of vehicular signals was measured and gathered in a multifaceted measurement in respect of the mentioned fleet. In Section 2, the measurement devices used and the sampling approach applied in this multifaceted measurement are described in some detail. A richer subset of this measurement data, including spatiotemporal and inertial data on stronger brakings, was evaluated from a road safety aspect in (Fazekas and Gaspar, 2010). More complete data sets could be used for continual monitoring of road safety of certain routes - in case of trucks running mostly over some standard routes - and also in the given truck route monitoring context.

Having considered the needs of companies with truck fleets of moderate size in respect of route monitoring, two preprocessing steps are proposed in the present paper in respect of the recorded trajectory and vehicle speed data. The first of these is segmentation of trajectories into meaningful trajectory sections, while the second is the automatic rendering of these sections - based on their spatiotemporal characteristics - to vehicle activity classes. These preprocessing steps can facilitate both the human and the computerized monitoring of truck routes, particularly if pointers to the detected activities within the trajectory data are used for indexing purposes.

The rest of the paper is structured as follows. In Section 3, Google Earth is used for the visualization of the route data converted into Keyhole Markup Language (KML). In Section 4, a linguistic approach for segmenting truck trajectories based on their spatiotemporal features is described. Some concrete trajectory sections, corresponding to certain low-speed maneuvers and stoppages, are looked at in detail. Then, a simple pictorial language for describing truck trajectories is proposed. In the last section, conclusions are drawn and further work pertaining to intelligent monitoring of truck trajectory data is suggested.

2. Measurement and sampling of truck positions and movements

Apart from a GPS sensor and an appropriate data communication device for sending the measured data to a dedicated data collection centre, several measurement devices were installed on trucks of the fleet mentioned in the introduction. These measurement devices included wheel-speed sensors, pressure sensors for measuring demanded and factual brake pressures, a steering wheel angle sensor and a yaw rate sensor.

The measurement of the vehicle speed was accomplished via measuring the angular velocities of the drive axle wheels and multiplying the measured values with appropriate calibration constants. Each speed sensor used in the measurements comprises a permanent magnet connected to a metal rod with a wire coil and a tone wheel mounted to the hub of the particular wheel. Voltage is induced in the sensor coil as the teeth pass the sensor head. The wheel-based vehicle speed is calculated from several angular velocity measurements of the wheels. Locally, the measured speed values are collected via the truck's CAN bus. This intra-vehicular communication system is based on the Controller Area Network (CAN) data link layer protocol (ISO, 2003). The pressure transducers used in the measurement were located close to the delivery ports of the foot brake valve to measure the driver demanded brake pressure of both front and rear brake circuits. The pressure information was used to control the brake pressure during an Electronic Stability Program intervention. An additional pressure transducer was used to measure the pressure in the bellows of the rear axle air suspension. The information was used to support the vehicle mass estimation. The yaw rate sensor used in the measurement comprises two bulk micro-machined out-of-phase oscillating masses. These masses support surface micro-machined accelerometers for measuring the Coriolis acceleration. Apart from this measurement, a surface micro-machined capacitive measuring element was used to measure the lateral acceleration of the vehicle. The steering wheel angle sensor is a self monitoring sensor that measures the steering angle. The output signals of the sensor are transmitted via CAN bus. The total range of the steering wheel angle is usually larger than the sensors measuring range; therefore, the out-of-range measurements must be somehow identified. This can be signaled, for example, by otherwise unexpected sign changes.

Relying on the measurement records from the aforementioned measurement devices sent over the CAN bus of the vehicle, two different data samplings were carried out: a) an on-going, relatively infrequent (2 sample/minute)

registration of the vehicle's geographical position, the covered total distance, the actual vehicle mass and the wheel-based vehicle speed; and b) an event-based (Anti-lock Braking System and Roll Stability Program activity based) registration of a richer set of measurement data.

3. Visualization of truck trajectory data

The first step in monitoring truck routes is visualization of the collected trajectory data. In our case, the trajectory data is only a subset of the data produced by the multifaceted measurement outlined in the previous section. To give a feel of the measured data, some visualization examples are included from the mentioned data set.

In Figs. 1.a-b, the route covered by a particular truck of the monitored fleet is shown without and with the locations of stronger brakings marked, respectively. The size of the markers in Fig. 1b indicates the magnitude of the intended braking force. The brakings data acquired and sampled as outlined in the previous section is included here for illustrative purposes only. The braking data was used neither in the segmentation of routes, nor for identifying the truck activities both presented in Section 4.

In Figs. 1c-d, routes covered by a small fleet of trucks - again without and with geographical locations of stronger brakings, respectively - are shown. The colors of the lines and the markers identify the individual trucks of the fleet. The temporal sampling rate of the position measurements in the multifaceted measurement outlined in Section 2 was chosen to ensure an acceptable spatial sampling frequency for truck monitoring purposes.

Figure 1: Routes covered by a small fleet of trucks — displayed with Google Earth — over a satellite image.

3.1. Visualization with Google Earth

The truck trajectory data - and if required other measurement data pertaining to haulages, e.g., braking data -must be converted into graphical overlay description before they can be visualized. When using Google Earth for visualization, the contents of the graphical overlay must be specified in Keyhole Markup Language (KML), an XML-like graphical overlay description language developed by Google. Figs. 1a-d and 2, as well as Table 1 present

truck trajectories in this manner. For more information on geographic visualization and KML, see (Wernecke, 2008). Google Earth has useful features for checking truck trajectories; it can zoom in on the individual routes and braking locations, furthermore, it can "fly over" a given trajectory and show it as if it were seen from an airplane. Google's StreetView is a very convenient tool to look around in a virtual manner at/from some far away publicly accessible locations. For truck route monitoring purposes, these locations tend to be various road locations (e.g., locations of slow truck movements and stoppages).

4. Spatiotemporal segmentation and description of truck trajectories

The verification of proper truck use should certainly include checking the routes that are either being covered, or at some stage had been covered by the trucks. The first approach is the online tracking of vehicles, such a system is described for example in (Luculescu and Enache, 2010), while the second one is offline - post-trip - monitoring of vehicles, such a system presented in (GuardMagic, 2011).

Table 1. Characterisation of road segments and their neighborhoods (with satellite images of concrete truck routes).

The routes covered during long-haul transportation assignments can be of considerable length, often reaching many hundreds of kilometers per assignment. Considering this substantial length and the time required to cover such a distance, furthermore considering the minimal spatial and temporal sampling frequencies for thorough inspection of routes, say 1 sample/km, and 1 sample/minute, respectively, it is clear that a substantial amount of trajectory data is generated from the measurements, transmitted, stored and inspected for each assignment. It could be then helpful to segment the trajectory covered by a truck into meaningful trajectory sections that are easy to handle and analyze. Each of these trajectory sections could be then assigned, based firstly on recorded truck position and speed data, secondly on transportation specific data, and thirdly on general road data, to some vehicle activity class. These classes could include various high-speed and low-speed runs (e.g., on motorways and narrow roads) and different kinds of stoppages (e.g., short stoppage at road-side resting places). The transportation specific data should include

geographic and descriptive data on various past and future truck destinations, on accommodations frequently used by the drivers of the company, and also on preferred refueling stations. The general road data should provide characterization of road segments and their transport related environment. The characterization could take the form suggested in (Fazekas and Gaspar, 2010). The road and neighborhood categories used there are shown in Table 1.

In case of a high-speed run, its duration, the covered distance, the vehicle's average speed, and fuel consumption, as well as, the geographic coordinates of the route section's starting and end-points are the most important features a haulage company would be interested in. In case of a low-speed maneuver, its spatiotemporal character, or in other words, the spatiotemporal morphology of the trajectory, could be also of interest as it might give a clue concerning the intentions of the truck driver. In case of a stoppage, the apparent purpose of the stoppages would be important to identify. For this purpose, the type of the location should be known. The proposed segmentation of truck trajectories could, on one hand, help detect certain infringements of traffic rules and identify certain suspect, or malevolent driver behaviors. On the other hand, assuming bona fide drivers and good-willing companies, the collected stoppage data could be utilized in structuring trips and deploying shipments in a way that takes individual driving and stopping habits and preferred stopping locations of the truck drivers into consideration.

4.1. The Pictorial Truck Action Description Language

Pictorial signs, such as traffic signs, road signs and pictograms, have been and still are the most important means to control road traffic and vehicle/driver behavior on roads (Wagner, 2006). The importance of these signs and of their standardization was recognized in the International Convention on Road Signs signed in Vienna in 1968. Since then many countries have joined the original signatories of the convention. Pictorial signs are used in many GPS-based applications facilitating road vehicle navigation (Skog and Händel, 2009). A complex system of intricate vehicle movement codes is used in describing details of road accidents (PIARC (2007). As signs like the ones mentioned above convey complex spatial, temporal, speed-related and behavioral information in an easy-to-grasp manner, defining a simple pictorial truck action description language for monitoring purposes was considered useful.

A spatiotemporal linguistic approach for modeling and describing vehicles moving over a road network was presen-ted in (Vazirgiannis and Wolfson, 2001). Their vehicle movement description language is rather complex and gene-ral in terms of supported road vehicle-types. Furthermore, it supports a wide range of database queries that are ire-levant in the present context. Another spatiotemporal database system that handles road vehicles' GPS positional data is used to answer relational queries (e.g., Vehicle A beingBehind Vehicle B) about pairs of vehicles (Hornsby and King, 2008). Such queries frequently occur when managing platoons of trucks on a multi-truck freight assignment. An algebraic approach was presented for spatiotemporal modeling of vehicles over a road network in (Wang et al., 2005). Recently, a trajectory pattern miner application was developed for handling and querying road network trajectory data using a spatiotemporal approach (Roh and Hwang, 2011).

For the description of truck maneuvers (e.g., loading/unloading at some business location) and stoppages (e.g., parking at a roadside resting place), a fairly simple intuitive pictorial language, namely the Pictorial Truck Action Description Language (PTADL) is proposed here. PTADL relies primarily on the spatial traits of the trajectory, though the speed is also taken into consideration. PTADL focuses on low- and zero-speed truck maneuvers as these are considered more critical from a vehicle and shipment security point of view. In the rest of this section, parallel to the analysis of a concrete fairly simple stoppage episode and some more complex maneuvers, some of the primitive driving actions of the driver - or in a truck-centered view, some of the primitive motion elements of the truck - and their respective aggregated driving actions/motion elements are introduced. The pictorial descriptions of these actions will be also presented.

4.2. Simple driving actions

A contiguous part of a trajectory - corresponding to, say, more than 5 minutes elapsed time - will be referred as trajectory section. In Fig. 2a, the spatial trajectory section of a particular truck is marked with orange line. A trajectory section can be partitioned into short trajectory segments comprising up to four position samples. Each of these trajectory segments corresponds to a primitive driving action (PDA) of the driver, or in a truck-centered view, to a primitive motion element (PME) of the truck. For convenience, consecutive PDA's /PME's of the same type

can be aggregated. As the resulting compound actions are still relatively basic, these will be referred as simple driving actions (SDA's) and simple motion elements (SME's). With reference to the driving episodes discussed below, the various SDA's / SME's and their pictorial signs are introduced.

4.3. Case studies

4.3.1. A parking episode at a roadside resting place

Having been driving his truck for some time, the Driver A stopped his vehicle at a road-side resting place. He stayed there for a few minutes and drove on. The spatiotemporal details of this parking episode were recorded in the truck data collected in conjunction with the truck fleet mentioned earlier.

S - / • # / - F

Figure 2: Trajectory of a truck parking at a roadside resting place displayed with Google Earth (a). The road and the entrance of the rest place displayed with Google StreetView. The string of pictorial signs, each denoting a simple driving action, summarizes the parking episode. Different views of the episode's spatiotemporal trajectory (d) and (e).

The spatial trajectory section corresponding to this parking episode is shown in Fig. 2a. The episode's spatiotemporal trajectory section is presented in Figs. 2d. Only the longitude, latitude and time axes are shown in the figure as the altitude values are assumed to be "on surface". The lack of considerable position noise during the stoppage indicates that only a relatively short time was spent in the parking place. This is in agreement with the time range appearing in the figure; the spatiotemporal trajectory is simple and easy to interpret; there is no sign of any complicated parking maneuvers. The sequence of SDA's / SME's for the parking episode is presented in Fig. 2c in the form of a pictorial string. The concrete SME's marked in Fig. 2.a are explained below together with their pictorial signs. The truck drew near the resting place with high speed (a). Speed higher than 20 km/h is indicated by dark green (■). The truck kept to the right lane of the main road and its forward movement was over 20 km/h; its movement was straight. The SME is denoted by ^^ . Then the truck entered into the lane that leads off the road

into the roadside resting place (b). That is, a significant lane-change occurred. The SME is denoted by /. In the meanwhile, the truck slowed down. The low-speed movements - that is movements with speed lower than 20 km/h - are marked with light green (■). The truck moved along the exit-lane into a resting place, that is, it turned slightly off from its earlier course to the right (c). The corresponding SME is denoted by . Having arrived to the resting place, the truck stopped (d). This SME is denoted by The truck stayed in the parking place for about 20 minutes (e). A stoppage SME longer than 3 minutes is denoted by $ . The truck then started to move again. It moved slowly forward and turned slightly left (f). This SME is denoted by The truck then returned to the main road by carrying out a significant lane-change / (g). Then the truck moved forward on the main road with higher than 20 km/h speed (h). As mentioned above, this SME is denoted by ^^ . In Fig. 2d, the pictorial signs are placed close to their trajectory segments. SME's b and g were omitted from the diagram as information on the lane structure and geometry would have been necessary to identify them.

4.3.2. Loading/unloading at a business location

The spatial trajectory of a truck during its loading/unloading at a distribution hub is shown in Fig. 3a and b in two different resolutions. The time spent at this business location was considerably longer than the duration of the parking episode discussed above. Also, the complexity of this maneuver was higher than that of the parking episode.

Figure 3: Spatial trajectory of a truck during its loading/unloading at a business location displayed with Google Earth in different resolutions. A minuscule detail of the trajectory (c) near and around the location identified by the diamond-shaped marker in central figure (b). The spatiotemporal trajectory section of the truck in the loading/unloading episode with SDA's.

The truck's spatiotemporal trajectory appears in Fig. 3d. It is presented in a general axonometric view selected for good visibility of the details. The pictorial signs were placed close to the corresponding trajectory segments. The siens which have not been used earlier are more or less self-explanatory, except for sign for haphazard movement

A minuscule detail of spatial trajectory near and around the geographic location identified by the diamond-shaped marker in Fig. 3b is shown in higher resolution in Fig. 3c. Looking at the figure, one can distinguish between the "loose knots" of the trajectory section corresponding to the truck's haphazard movements and the "tight knots" which correspond to GPS position errors during stoppage. To differentiate between the haphazard movements and the random position noise, criteria concerning the amplitudes of the movements were set up. These are discussed later in the text.

4.3.3. A refueling episode

The trajectory of a truck during a detour to a petrol station is shown in Fig. 4.a. The morphological character of the maneuver is similar to that of some loading/unloading episode. In Fig. 4b, the SDA's are shown for the detour.

The business location involved here is a petrol station; so it is quite possible - and depending on the result of a detailed analysis of the spatiotemporal trajectory section over the site, it is quite likely - that the truck has stopped for the refueling. It normally takes considerably less time than loading/unloading a truck, but of course petrol stations also function as rest places, cafes and shops, and so the time spent there is not necessary spent only for truck refueling.

Figure 4: The trajectory section of a truck near a petrol station during a refueling episode displayed with Google Earth (a) and the corresponding simple/compound driving actions (b).

4.4. Dimensional, temporal and other criteria for detecting maneuvers

As stated earlier, the PDA's / PME's are determined from four sample points for the temporal sampling rate used in the paper. If other sampling rate is used then the number of samples per PDA should be chosen to keep the PDA's duration approximately the same. Consecutive PDA's of the same sort are aggregated into SDA's.

In order to further compress the description, simple grammatical rules can be used to aggregate different PDA's in some acceptable and useful manner, e.g., multiple slow and short movements could be seen as a compound driving action (CDA); e.g., consecutive slow/fast haphazard movements form a CDA.

Truck maneuvers are even more complex patterns of movements, or in the driver-centered approach, even more complex sequences of driving actions, carried out usually with a clear purpose. Typical characteristics of maneuvers can be collected and can be described in the spatiotemporal, geographic and other domains.

Figure 5: Detection of a refueling maneuver. The petrol station of Fig. 9.a displayed with Google StreetView (inlay).

Now the processing involved in the detection of slow and zero-speed maneuvers is outlined. First, only the relevant SDA's are considered. These are the directional changes and the significant lane-changes. If several relevant SDA's are identified in close spatiotemporal vicinity, then the subordinate driving actions, namely the haphazard movements, short and long stoppages and, are sought within some predefined distance around the geographical points corresponding to the detected relevant SDA's.

Fig. 5 illustrates the maneuver detection in respect of the trajectory section shown in Fig. 4a. The SDA's and the CDA's corresponding to this trajectory section are shown in Fig. 4b. In Fig. 5, the relevant SDA's are located first and are used for further processing. The detected relevant SDA's are as follows: turning left (blue arrow), turning slightly left SDA (green arrow), turning slightly right (orange arrow), and turning slightly left (yellow arrow).

The search areas for locating subordinate driving actions are shown in matching colors. These disks have the same predefined radii and are drawn around the geographic locations corresponding to the four relevant SDA's detected.

Within the search disks subordinate driving actions are sought and three such actions are identified. The corresponding signs are shown in light gray. If subordinate driving actions are present in the search area, as is the case for Fig. 5, then these driving actions are checked for their duration, their speed range and the displacement they cause The white circle in Fig. 5 circumscribes the relevant SDA's and the subordinate driving actions; it marks the approximate location of the detected maneuver, in this case, refueling.

5. Conclusions and further work

To facilitate intelligent monitoring of routes and stoppages of trucks from recorded route data, particularly in case of relatively small transportation companies, two preprocessing steps were proposed. These steps are the segmentation of trajectories into meaningful trajectory sections and the automatic rendering of route sections to vehicle activity classes. The trajectory segmentation method is based on spatiotemporal features of the trajectory. Then the trajectory and speed data are looked at and processed at two different semantic levels. The detected maneuvers and their pointers within the trajectory data can be used for indexing purposes to facilitate the route monitoring. The trajectory sections were displayed as graphical overlays on satellite images. In the paper, more emphasis was given to the low- and zero-speed maneuvers as these are more critical from a shipment security point of view. A simple spatio-temporal pictorial description language was proposed. A pilot study is on its way to construct reliably performing PTADL grammars and useable dimensional constraints for detecting various maneuvers.

Acknowledgements

The financial support provided by the Hungarian National Office for Research and Technology - in the frame of grant TECH-08-2/2-2008-0088 - is gratefully acknowledged.

References

Fazekas Z. and Gâspâr P. (2010). Visualisation and safety evaluation of truck route and braking data. Proc. of 11th Int. Congress on Automotive and Transport Engineering, Brasov, Romania, pp. 11-16.

GuardMagic (2011). Vehicle off-line GPS and fuel monitoring. http://www.guardmagic.com/01-engl/2e-products/e-doc/04GuardMagic.off-line monitoring.pdf, GuardMagic SIA, Riga, Latvia.

Hornsby, K. S. and King, K. (2008). Modeling motion relations for moving objects on road networks. Geoinformatica vol. 12, pp. 477-495.

ISO (2003). Road vehicles — Controller area network (CAN) — Part 1: Data link layer and physical signaling. ISO 11898-1:2003.

Luculescu, M. C. and Enache, C. J. (2010). Digital system for monitoring andmanagement of individual or fleet automotive vehicles. 11th International Congress on Automotive and Transport Engineering, Brasov, Romania, pp. 143-150.

PIARC (2007). Road accident investigations guidelines for road engineers. World Road Association.

Roh, G-P., Hwang, S-W. (2011). TPM: Supporting pattern matching queries for road-network trajectory data. Proc. 14th Int. Conf. on Extending Database Technology, Uppsala, Sweden, pp. 554-557.

Skog, I. and Händel, P. (2009). In-car positioning and navigation technologies: a survey. IEEE Transactions on Intelligent Transportation Systems, vol. 10, pp. 4-21.

Vazirgiannis, M. and Wolfson, O. (2001). A spatiotemporal model and language for moving objects on road networks. In: Advances in Spatial and Temporal Databases, Jensen, C. S. et al. (Eds.), Lecture Notes in Computer Science, Springer-Verlag, vol. 2121, pp. 20-35.

Wagner, A. (2006). The rules of the road, a universal visual semiotics. International. Journal for the Semiotics of Law, vol. 19, pp. 311-324.

Wang, S-S.. Liu, D-Y.and Liu, J. (2005). A new spatial algebra for road network moving objects. International Journal of Information Technology, vol. 11, pp. 47-58.

Wernecke, J. (2008). The KML handbook: Geographic visualization for the Web, Addison-Wesley Professional.