Vehicle Detection and Scale Adaptive Tracking Using Tail Light Segmentation

This paper proposes a vehicle detection and tracking system based on forward looking CCD camera. Vehicle tail light location information is employed to generate vehicle candidate, and tail light pair distance is used to adjust vehicle tracking window size. In vehicle detection step, it uses a back propagation neural network (BPNN) which is trained by Gabor feature set. BPNN verifies vehicle candidates and ensures system robustness. In vehicle tracking step, it also uses mean shift tracking algorithm using color feature space. In the experiment, 104 images in total are tested for vehicle detection, 87 vehicle images are detected successfully; a 175 frame video is tested for real-time vehicle tracking. These results show that the proposed system is effective for vehicle detection and tracking in the day time. Keywords-vehicle detection; vehicle tracking; color segmentation; mean shift