Multi-objects tracking and online identification based on SIFT

This paper presents a method of point feature tracking and online identification using SIFT(Scale Invariant Feature Transform). The proposed approach uses the probabilistic voting method with appearance model to estimate the object´s optimal center and apply hierarchical vocabulary tree to recognize the object. Since SIFT feature is invariant to changes caused by rotation, scaling and illumination, we can obtain higher tracking performance than the conventional approach and the probabilistic voting approach enables the track to search object efficiently. Online identification is also a challenge in video surveillance system, we use bag of words method based on hierarchical vocabulary tree to represent and match tracked objects by sampling SIFT feature online. Experimental results illustrate that the proposed approach works robustly for multi-persons tracking and identification.