296
Alminas ˇ
Civilis, Christian S. Jensen, and Stardas Pakalnis
application of SSC, TSC, and DSC. Vector-based tracking is also included. To avoid
clutter, only the most interesting techniques are illustrated for the AKTA data.
In the comparison, 568,307 GPS records were used from the INFATI data set and
about 4,000,000 GPS records from the AKTA data set. The curves show experimental
results using thresholds ranging from 100 to 1,000 m.
All three road network modifications increase the performance of segment-based
tracking, which then outperforms vector-based tracking. Segment-based tracking has
the best performance when using the road network resulting from the direction-based
modification.
The performance of a theoretical form of tracking that is optimal under the assumption that the speed of an object is modeled as being constant between updates
is also included in Fig. 13.4. This technique is explained in Sect. 13.5.
13.5 Update Reduction Using Routes
The focus of this section is on the use of the routes of moving objects for update
reduction. We first describe the theoretical, constant-speed optimal form of tracking
mentioned in the previous section. Then we consider the use of an object’s routes,
which are ‘long’ segments, during segment-based tracking instead of the previous
use of road-network segments.
13.5.1 Theoretical, Constant-speed Optimal Tracking
One may distinguish between updates based on the outcomes of the associated map
matching. Recall that in segment-based tracking, when the server receives an update
at a position p i , it attempts to map match the position onto the road network rn to
find the most probable polyline mpl and point mp on it.
Let MM be the map matching function and MM(p i , rn) = (mpl, mp). If MM
fails to identify a polyline and point, tracking is done in vector mode. Assuming that
the map matching is successful and
MM(p i−1 , rn)
.mpl =
MM(p i , rn)
.mpl, where
position p i−1 is that of the previous update, we say that the update is caused by speed.
If the polylines differ, we say that the update is caused by a segment change.
The theoretical, constant-speed optimal tracking introduced here indicates how
it is possible to achieve few updates with segment-based tracking in the best case,
which occurs when a moving object travels on only one segment and no updates
occur due to segment changes. The technique is optimal under the assumption that
the speed of an object is modeled as being constant between updates.
This technique is interesting because it offers a measure of optimality. However,
the technique is useful for comparison purposes only and it is not a practical technique. The technique is impractical because it assumes that the entire polyline along
which a vehicle will ever move is known in advance. We are able to use this technique here because we have the entire GPS logs for each vehicle. Using these, we
simply construct (very long) polylines that precisely track each vehicle “ahead of
time.” In practice, GPS positions are received in real time.
Alminas ˇ
Civilis, Christian S. Jensen, and Stardas Pakalnis
application of SSC, TSC, and DSC. Vector-based tracking is also included. To avoid
clutter, only the most interesting techniques are illustrated for the AKTA data.
In the comparison, 568,307 GPS records were used from the INFATI data set and
about 4,000,000 GPS records from the AKTA data set. The curves show experimental
results using thresholds ranging from 100 to 1,000 m.
All three road network modifications increase the performance of segment-based
tracking, which then outperforms vector-based tracking. Segment-based tracking has
the best performance when using the road network resulting from the direction-based
modification.
The performance of a theoretical form of tracking that is optimal under the assumption that the speed of an object is modeled as being constant between updates
is also included in Fig. 13.4. This technique is explained in Sect. 13.5.
13.5 Update Reduction Using Routes
The focus of this section is on the use of the routes of moving objects for update
reduction. We first describe the theoretical, constant-speed optimal form of tracking
mentioned in the previous section. Then we consider the use of an object’s routes,
which are ‘long’ segments, during segment-based tracking instead of the previous
use of road-network segments.
13.5.1 Theoretical, Constant-speed Optimal Tracking
One may distinguish between updates based on the outcomes of the associated map
matching. Recall that in segment-based tracking, when the server receives an update
at a position p i , it attempts to map match the position onto the road network rn to
find the most probable polyline mpl and point mp on it.
Let MM be the map matching function and MM(p i , rn) = (mpl, mp). If MM
fails to identify a polyline and point, tracking is done in vector mode. Assuming that
the map matching is successful and
MM(p i−1 , rn)
.mpl =
MM(p i , rn)
.mpl, where
position p i−1 is that of the previous update, we say that the update is caused by speed.
If the polylines differ, we say that the update is caused by a segment change.
The theoretical, constant-speed optimal tracking introduced here indicates how
it is possible to achieve few updates with segment-based tracking in the best case,
which occurs when a moving object travels on only one segment and no updates
occur due to segment changes. The technique is optimal under the assumption that
the speed of an object is modeled as being constant between updates.
This technique is interesting because it offers a measure of optimality. However,
the technique is useful for comparison purposes only and it is not a practical technique. The technique is impractical because it assumes that the entire polyline along
which a vehicle will ever move is known in advance. We are able to use this technique here because we have the entire GPS logs for each vehicle. Using these, we
simply construct (very long) polylines that precisely track each vehicle “ahead of
time.” In practice, GPS positions are received in real time.