Onboard vehicle detection and tracking using boosted Gabor descriptor and sparse representation

Abstract

Proposed is a new onboard vehicle detection method based on a part-based model. It uses several boosted Gabor descriptors of keypoints to represent the vehicle. To perform detection, the sparse representation-based classifier is adopted to classify the extracted keypoints in video frames. Then, by using the K-means algorithm, vehicle candidates with high-density classified keypoints are generated. With the keypoint matching adopted, these candidates can be verified, and the matched pairs are meanwhile to be used for vehicle tracking. Experimental results show that the proposed method is robust to environmental changes as well as achieving high detection accuracy.