Research Of Object Tracking Based On Multi-kcf

Online visual tracking of an arbitrary temporally changing object,specified at the first frame,has been a hot topic in computer vision for the last decades.And it still remains very challenging due to practical factors like illumination variation,background clutter,occlusions,scale variation,fast motion and deformation.And high performance visual tracking algorithms with good tracking accuracy and efficiency are required by numerous applications like video surveillance,traffic monitoring,autonomous systems,vehicle navigation,human-computer interaction,and video editing,to name a few.Correlation filter has been a widely used framework for visual object tracking,due to their superior computation and fair robustness to photometric and geometric variations.However,inappropriate model update will lead to tracking drift when the tracked object is occluded or changes in scale.In this paper,we propose a multi-KCF(kernelized correlation filter)tracking framework that integrates four KCF models based on histogram of oriented gradient(HOG)features.These four models include one traditional KCF tracker,one KCF tracker to detect whether the object is occluded,one KCF tracker to re-detect the tracked object after occluded,one KCF model to conduct scale estimation.During tracking,the traditional KCF tracker will estimate the position of target,meanwhile the KCF tracker based on target's appearance model will judge whether the target is occluded,which infects the update of model.When the target is occluded,the re-detect KCF will re-detect the area near the target in the previous frame to relocate the target.And we use classic pyramid scale estimate algorithm to deal with the scale variation.For estimating the performance of our algorithm,we compare with nine classic trackers in Object Tracking Benchmark.The experimental results demonstrate that the proposed method achieves better performance than some state-of-the-art trackers in terms of both accuracy and robustness,especially in dealing with occlusions.