Object tracking based on ORB and temporal-spacial constraint

Object tracking is an important issue in video surveillance. In this paper, we present a tracking framework based on ORB (Oriented FAST and Rotated BRIEF) feature using temporal-spacial constraint. ORB is a fast binary descriptor which needs low computation cost and has similar matching performance with SIFT or SURF. Firstly, ORB keypoints and their descriptors of the object are calculated on two adjacent frames. Afterwards, a best match between these two point sets is obtained through computing Hamming distance. Then, the next location of the object can be deduced from the matched keypoints. In the case of without match, we can still predict the target's location using temporal-spacial constraint. The performance of our tracking system is evaluated on datasets of real surveillance scenes. The experiments results show that the presented approach can track objects accurately and efficiently.