Target Contour Recovering for Tracking People in Complex Environments

Figure 12

Examples of occlusion detection and target shape recovering in real-world video sequences. The frame numbers from top to bottom are 1983, 1989, 2027, and 2043, respectively. The columns from left to right are original frames, foreground contours, results of occlusion detection, and results of shape recovering, respectively. In the second column, we have eliminated other foreground contours which are not connected to our target for clearer display. Note that the results of occlusion detection are quite exact but the result of shape recovering in the third row is failure.