A Learning Based Approach for Vehicle Detection

This paper presents a uniform learning based method for detecting different types of vehicles from images captured on moving vehicles. Learning based methods have shown great robustness in vehicle detection but are restricted by the variety of vehicle types, which leads to different appearance in images. The common parts of all vehicles are their bottoms: mainly two wheels and a shadow. In our approach, all types of vehicles are first detected by their bottoms. Symmetry and edge detections are then performed to obtain the exact positions of the vehicles. Experimental results show the efficiency of our approach. Our approach even outperforms a specific sedan classifier when detecting sedans