The objective of this project is to develop a methodology for generating traffic statistics from high-resolution satellite images. Most current systems for obtaining traffic statistics are based on systems located in the road or very close to the road. Such systems, typically induction loops, count vehicles that pass a given point in the road system over time. In cases were the traffic distribution in an entire road network is of interest this methodology has evident shortcomings due to the very limited geographical coverage of such a system. The price of covering the entire road network with such counters would be “astronomical” and alternatives are necessary.

 

High-resolution satellite images provide a very interesting source of information for this application. The resolution of the current best optical satellites (roughly 0.5 m) is such that vehicles are clearly distinguishable. In a project financed by ESA, we are developing a methodology for counting vehicles in such images. Our concept is a two-step approach relying on first segmenting the vehicles to generate candidate vehicles and then to classify these candidates. Current results are promising and we have good hopes of being able to generate relevant traffic data based on such images.