292
Alminas ˇ
Civilis, Christian S. Jensen, and Stardas Pakalnis
The results were obtained by simulating the scenario described in Sect. 13.2.1
with thresholds ranging from 100 to 1,000 m. Specifically, the movement of each
car was simulated using the log of GPS positions for the car. So a client program and a server program interact, and a simple experiment management system is in charge of the bookkeeping needed to obtain the performance results.
Instead of obtaining GPS positions from a GPS device in real time, the client program utilizes the GPS logs, which of course makes the simulation much faster
than the reality being simulated. The bookkeeping involves the counting of updates sent from the client program to the server program and keeping track of
time.
All performance studies reported in this chapter follow this pattern. The studies
differ in the specific GPS data and road networks used, and in the tracking policies
used.
In Fig. 13.2, accuracy threshold values in meters are on the x-axis. The client
obtains a GPS position from the GPS device every second and performs a comparison between the GPS position and the predicted position. The y-axis then gives the
average number of seconds between consecutive updates sent from the client to the
server to maintain the required accuracy.
It is seen that the time between updates increases as the accuracy threshold increases, that is, as the required accuracy decreases. Point-based tracking shows the
worst performance. The largest improvement of the segment- and vector-based techniques over the point-based technique is for smaller thresholds, while for larger
thresholds the improvement is smaller. For thresholds below 200 m, segment- and
vector-based tracking policies are more than two times better than point-based
tracking.
Segment-based tracking is outperformed because the road segments in the underlying road network are relatively short, having an average length of 174 m. For
example, this means that a relatively straight road is represented by several segments.
In this case, vector-based tracking may need less updates. So, although vector-based
tracking is simpler and performs slightly better, we find it likely that it is possible to
improve segment-based tracking to become the best.
In addition, segment-based tracking, by relating the location of a moving object
to the underlying road network, offers other advantages that are as follows:
• Buildings, parking places, traffic jams, points of interest, traffic signs, and other
road-related information that is mapped to the road network can easily be associated with the location of a moving object.
• Road network-based distances can be used in place of Euclidean distances.
• Acceleration profiles, driver behavior on crossroads, and other road-related data
that increase the knowledge about the future positions of moving objects can be
exploited.
Consequently, a promising direction for obtaining improved tracking is to continue in the direction of segment-based tracking.
Alminas ˇ
Civilis, Christian S. Jensen, and Stardas Pakalnis
The results were obtained by simulating the scenario described in Sect. 13.2.1
with thresholds ranging from 100 to 1,000 m. Specifically, the movement of each
car was simulated using the log of GPS positions for the car. So a client program and a server program interact, and a simple experiment management system is in charge of the bookkeeping needed to obtain the performance results.
Instead of obtaining GPS positions from a GPS device in real time, the client program utilizes the GPS logs, which of course makes the simulation much faster
than the reality being simulated. The bookkeeping involves the counting of updates sent from the client program to the server program and keeping track of
time.
All performance studies reported in this chapter follow this pattern. The studies
differ in the specific GPS data and road networks used, and in the tracking policies
used.
In Fig. 13.2, accuracy threshold values in meters are on the x-axis. The client
obtains a GPS position from the GPS device every second and performs a comparison between the GPS position and the predicted position. The y-axis then gives the
average number of seconds between consecutive updates sent from the client to the
server to maintain the required accuracy.
It is seen that the time between updates increases as the accuracy threshold increases, that is, as the required accuracy decreases. Point-based tracking shows the
worst performance. The largest improvement of the segment- and vector-based techniques over the point-based technique is for smaller thresholds, while for larger
thresholds the improvement is smaller. For thresholds below 200 m, segment- and
vector-based tracking policies are more than two times better than point-based
tracking.
Segment-based tracking is outperformed because the road segments in the underlying road network are relatively short, having an average length of 174 m. For
example, this means that a relatively straight road is represented by several segments.
In this case, vector-based tracking may need less updates. So, although vector-based
tracking is simpler and performs slightly better, we find it likely that it is possible to
improve segment-based tracking to become the best.
In addition, segment-based tracking, by relating the location of a moving object
to the underlying road network, offers other advantages that are as follows:
• Buildings, parking places, traffic jams, points of interest, traffic signs, and other
road-related information that is mapped to the road network can easily be associated with the location of a moving object.
• Road network-based distances can be used in place of Euclidean distances.
• Acceleration profiles, driver behavior on crossroads, and other road-related data
that increase the knowledge about the future positions of moving objects can be
exploited.
Consequently, a promising direction for obtaining improved tracking is to continue in the direction of segment-based tracking.