Tracking a Variable Number of Human Groups in Video Using Probability Hypothesis Density

We apply a multi-target recursive Bayes filter, the probability hypothesis density (PHD) filter, to a visual tracking problem: tracking a variable number of human groups in video. First, we use background subtraction to detect human groups which appear as foreground blobs. The PHD filter is implemented using sequential Monte Carlo methods; and the centroids of the foreground blobs are used as the measurements to update the PHD filter. Our experimental results show that when human groups appear, merge, split, and disappear in the field of view of a camera, our method can track them correctly