Real-Time Object Tracking Using Powell's Direct Set Method for Object Localization and Kalman Filter for Occlusion Handling

摘要

A new kernel-based method for real-time tracking of objects seen from a static or moving camera is proposed. The central processing block uses Powell's direct set method to optimally find the most likely target position in every frame. The changes in object shape, scale, orientation and shading conditions have been dealt by a template adaption module which exponentially forgets the past features of object and incorporates latest feature into template after every frame. The proposed algorithm also handles short-term partial and full occlusion by using Kalman filter for trajectory prediction and proximity search for relocking object once it reappears in the scene. The experimental results show robust tracking of a variety of objects undergoing severe occlusion and significant appearance changes, with an extremely low computational complexity. The proposed tracker can perform tracking at an average frame rate of 60 frames/sec, which is sufficient for real-time applications.