Estimation of traffic parameters in urban areas from satellite images

In this paper an approach for automatic vehicle detection from satellite imagery is presented. In space-borne images single vehicles have a very limited size and a detection scheme which only focuses on single vehicles would deliver ambiguous results. Therefore we extract vehicle queues because their repetitive pattern improves the distinction from similar objects. Additionally information about the location and approximately size of roads are used to limit the search space and thus decrease the number of false detection. Single vehicles can be determined by analysis of the width and contrast behavior of the queues. The results show that the analysis of width and contrast information extracts vehicles with high reliability, but since only a certain number of vehicles appear grouped, the overall completeness is not sufficient yet. The used data sets are Quick Bird images covering the inner city of Munich (Germany) with a ground sampling distance of approximately 0.6 m for the panchromatic channel.