Ghost, occlusion and distractors handling in object tracking system

For a real time object tracking system, potential interferences such as ghost, occlusion and distractors have a great influence on the tracking performance. This paper aims to provide a solution to these problems. In order to remove the ghost, we first narrow the interest region of the foreground to the boundary region of the tracked object. Then the new interest region on the incoming image normally contains small amount of contour information if the foreground is a ghost. This can be used for fast ghost removal. As for the occlusion handling, we adopt the merge/split strategy as well as take advantage of the hue histogram in HSV space to distinguish the merged objects. In addition, the main moving direction of the tracked object will be obtained in advance for detecting the distractors case. As long as such case happens, the hue histogram is used to identify whether the tracked object is the original one so as to avoid the distractors problem. We have applied our algorithm to many banking supervision related videos. The experimental results show that the proposed tracking scheme is feasible and competent.