CAMShift-Based Moving Object Tracking System

Abstract

Moving object tracking is the process of locating a moving target over time in sequential video frames, and it is an essential research topic in the field of computer vision. As a well-established and efficient algorithm for kernel-based object tracking, CAMShift performs well with simple moving objects, but it is not robust in more complex situations. To overcome the defects in the process of tracking moving objects, a combination of CAMShift and Kalman Filter is widely used in industrial applications and practices. A CAMShift-based tracking system with Kalman Filter-based track window prediction is presented in this paper. When the camera is moving or the object is occluded, the velocity of the moving object is used by Kalman Filter to provide a prediction of the tracking window. Both the original CAMShift-based tracker and the hybrid system are evaluated by the LaSOT benchmark. The results show that the hybrid tracking system has a better success rate on certain categories of objects and situations than the original CAMShift tracker.