Key points are not available for this paper at this time.
Algorithms of video traffic image processing, presented below, have been developed by CMM and INRETS using 256x256x6 bits images collected from various scenes covering a 150 m area of a freeway under various weather conditions. First, road detection is automatically performed. The road and traffic lanes images are used to derive relationships between real and image distances and to build image transformations independant from the perspective view. Then, markers of the vehicles are extracted using geometrical adapted filters. These markers are joined to define a single marker for each vehicle. Finally, vehicles trajectories are built and traffic variables are specified.
Beucher et al. (Fri,) studied this question.