Position analysis based on multiple views in multi-target tracking

Abstract

A spatial position analysis algorithm for multi-target tracking is proposed in multiple views. A target centroid marked is obtained from the zero and the first moments. Data association between a centroid and a trajectory exploits fuzzy C-mean clustering algorithm. Positions of moving targets are calculated by triangle principles in stereo vision. Furthermore, back projections from a three-dimensional point to a camera plane in each view angle are compared to the centroid in order to obtain RMS error. Experimental results indicate the method can reduce computational complexity and also be robust to occlusion, which can meet actual needs.