Real-time GPU implementation of an improved cars, pedestrians and bicycles detection and classification system

In this paper a real time system for cars, pedestrians and bicycle detection and classification is presented. The system aims at monitoring the traffic flow in urban zones and uses video data acquired with both mono and stereo cameras. All the algorithms have been developed in a pixel-wise fashion in order to be parallelized on a GPU device for real-time performances. We show that by a GPU implementation of the time-consuming parts of the proposed system, we perform detection and classification at about 25 frame per second to ensure prompt and effective reaction to the monitored events.