Visual tracking tracker via object proposals and co-trained kernelized correlation filters

Abstract

Visual tracking is a challenging task in the field of computer vision with wide applications in intelligent and surveillance systems. Recently, correlation trackers have shown great achievement in visual tracking due to its high efficiency. However, such trackers have a problem of handling fast motion, motion blur, illumination variations, background clutter and drifting away caused by occlusion and thus may result in tracking failure. To solve this problem, we propose a tracker that is based on the object proposals and co-kernelized correlation filters (Co-KCF). The proposed tracker utilizes both object proposals and global prediction estimated by kernelized correlation filter scheme to obtain best proposals as prior information using spatial weight strategy in order to improve tracking performance of fast motion and motion blur. Since single kernel may lead to background clutter and drifting problem, Co-KCF has been employed to combat this defect and predict a new state of a target object. Extensive experiments demonstrate that our proposed tracker outperforms other existing state-of-the-art trackers.