A new approach to detect pedestrian vehicles and bicycles

Multi-view approach has been proposed in traffic monitoring for its robustness. Though there is much research work in classification, it is not easy to do so. In this paper, we describe a new approach to detect the pedestrian, vehicle and bicycles in the traffic relying on the muti-view information. In this approach, we do not detect objects from any single view; information is gathered from all of the views into an integrated framework and detection results are transformed to one view. Without relying on calibrated views, we use only 2D constructs to do the work. To this end, we adopt homographic constraint to obtain the synthesized foreground above the ground from multiple views. For the classification between pedestrian, vehicle and bicycles, we use SVM classifier base on the shape and size information. Experiment shows that our algorithm can classify pedestrian, vehicles and bicycles more accurate and easier and it provides a new way to use the muti-view information.