Tracking Failure Detection Using Time Reverse Distance Error for Human Tracking

摘要

This paper proposes a tracking failure detection method based on time reverse error distance. Corner feature is used as interest point to perform the task. The algorithm consists of three stages. First, the corner feature is extracted from the image. Then, a tracker produces a trajectory by tracking the point in the previous frame to the current frame. Second, the location of point in the current frame is initialized as a reference point. The validated trajectory is obtained by tracking-reverse the reference point in the current frame to the previous frame. Third, both trajectories are compared with each other, if they are significantly different, the reversed point is considered as an incorrect. To evaluate the performance of this method, an object tracking method is performed. A set of points is initialized from a rectangular bounding box. These points are tracked and evaluated using the proposed tracking failure detection method. The correct tracked points are used to estimate bounding boxes in consecutive images. The performance results show that the proposed method is efficient for human detection and tracking in omnidirectional images.