On-road vehicle detection method based on multi-scale active basis model

This paper introduces a vehicle detection method based on multi-scale active basis model in traffic surveillance systems. Due to the effects of perspective, vehicles which are close to the camera are larger and more detailed than the far ones on individual video frames. Using camera calibration, we get the multi-scale information of vehicles, and then we learn the multi-scale active basis model from the target training sample set by using the shared sketch algorithm. The multi-scale ABM can detect vehicles in various poses, shapes, and sizes in a whole video frame. The experiment results show that this proposed method can fit the changes of vehicle size in images.