A multi-view object tracking using triplet model

Abstract Object tracking is very important in intelligent systems, such as video surveillance, automatic drive and traffic security. Background subtraction algorithm is a mature method for foreground object extraction, but it may be affected by complicated background or the change of object shape. So in this paper, a novel object tracking method is proposed using a triplet model. First, BING feature is used to find some potential object proposals. Then, we construct a triplet model for each potential object. The triplet of the same object between two consecutive frames is considered similar. Finally, object tracking can be achieved by computing feature difference of triplets. Experimental results show that our method can achieve object tracking effectively and in real time.