A FRAMEWORK FOR ROBUST OBJECT TRACKING BY COMBINING MULTIPLE OPENCV ALGORITHMS

This paper describes a framework for robust object tracking by combining the results of several tracking algorithms. Object tracking is complicated by various factors, such as changes in the shape and/or movement direction of the tracked target, and occlusion by other objects. Object tracking algorithms each have individual strengths and weaknesses, and even an algorithm that is highly accurate in some circumstances may be less accurate in others. In the framework proposed here, the tracking results of several algorithms provided in OpenCV are integrated into a more reliable tracking result based on the centroid computed from a weighted average of center points determined by each individual algorithm. The results of experiments using video sequences demonstrate that the proposed framework was able to track a target object successfully even in cases in which most of the OpenCV algorithms individually failed to do so.