Research on target association based on feature extraction

In non-overlapping surveillance camera network, the observations of targets are often widely separated in time and space. Therefore target association becomes a key problem to realize target tracking. In this paper, let pedestrian be research target, a feature extraction model based on observation's main color statistical information and spatial distribution information is put forward. Then, similarity calculation method is proposed. And by setting the threshold to determine whether the two targets' are associated. Experimental results show that our method can achieve reasonable similarity value and matching result, and have good robustness to the change of shape and gesture.