Tracking partially occluded objects with centripetal active contour

Abstract

Abstract. This study develops a new approach for tracking single object through a video in case of partial occlusion. The approach uses an active contour (AC) to determine the mass center of the target in the present frame. The mass center is utilized by a Modified Kalman Filter which estimates its position, in the next frame, as the center of the area, the target is most likely to appear in. This area is bounded by the AC, which evolves toward the target. If multiple objects are enveloped, a shell algorithm splits the AC to multiple contours utilizing the mass center position estimated in the previous frame. Every shell is constructed by a part of the AC which envelops the visible, from the contour, parts of objects and ends at the mass center of the AC. In every shell, an AC extracts the object's boundary and generates a finite numerical sequence as a signature for this boundary. A sequence alignment approach matches the signature of every object, in the present frame, with the signature of the target. To validate the method, experiments are performed on 14 videos. The paper ends with discussion on the results and the contributions comparing them with contemporary works.