The International Archives ofthe Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
559 
Figure 3. Example for road extraction at A96 exit Munich-Blumenau (clipping from nadir image). Upper panel shows line 
detections at a flight height of 1000 m, panel below shows the resulting road area after smoothing / gap filling. 
since the module for roadside marking detection finds the 
dashed midline markings and stores them in a separate class. 
In a next step, the roadside identification module, again with the 
help of the road database tries to correct possible errors (gaps 
and bumps) that might have crept in during the feature 
extraction phase. Furthermore, it smoothes the sometimes curly 
road boundary detections from feature extraction (see fig. 3). 
Gaps due to occlusion of the road surface by crossing bridges 
are closed, if gapping is not too large. This has the advantage 
that the course of the road is not lost, although the road itself is 
not seen at this place. However, it could lead to false alarms in 
the car detection. If cars are crossing the bridge, they might be 
assigned belonging to the occluded road below the bridge 
spuriously in car detection. However, we try to sort them out by 
alignment, since they are elongated perpendicular to the course 
on the occluded road. 
Figure 4. Implemented processing chain for a knowledge 
based road extraction, vehicle detection, and vehicle 
tracking on an image sequence. 
3.2 Vehicle Detection 
With the information of the roadside obtained in the processing 
step described before, it is possible to restrict vehicle detections 
and tracking only to the well determined road areas. This 
increases performance and enhances the accuracy of vehicle 
detection. Based on this, we developed an algorithm for the 
detection of vehicles which is described in the following. 
With the information about the alignment and direction of the 
roadside, we are able to mark all pixels belonging to the road. 
Thereby, the local road direction is included as an extra 
parameter into the marker of each pixel. For the vehicle 
detection, a Canny edge algorithm (Canny, 1986) is applied and 
a histogram on the edge steepness is calculated. Then, a k- 
means algorithm is used to split edge steepness statistics into 
three parts which represent three main classes. These three 
classes are namely edges belonging to vehicles, edges 
belonging to roads, and edges within road and vehicle edges, 
and therefore not yet classifiable. 
We consider the part with the lowest steepness in the edge 
histogram being mainly populated by pixels of the road 
background, since its intensity is quite uniform. Moreover, we 
assume that the part with the highest steepness -due to the high 
discontinuity in the intensity- is most likely populated by 
vehicles. However, this part of the statistic is not only occupied 
by vehicle edges. It can be also contaminated by midline 
markings, shadows, sign boards, trees, or the like. 
In the histogram part containing the edges not yet classified, it 
must be determined which pixels belong to the road background 
or to potential vehicles. For this decision, the pixel 
neighbourhood is examined. Pixels directly connected with a 
potential vehicle pixel are moved into the vehicle class. 
Remaining pixels are finally considered as road background and 
neglected.