Research On Target Tracking Algorithm Based On Correlation Filtering

With the increasing demand of people to daily life,object tracking as an important research topic in the field of computer vision has been gradually favored by researchers.Target tracking technology can be applied to many aspects in real life,such as intelligent monitoring,intelligent transportation,human-computer interaction,medical diagnosis and so on.In recent years,target tracking algorithms based on correlation filtering have been proposed continuously.These algorithms have a very good effect in terms of accuracy and speed,but there are still many problems to be solved urgently.In order to solve problems such as scale change and occlusion,this paper selects two algorithms for improvement among various kinds of correlation filtering algorithms.The main research contents of this paper are as follows:(1)In the tracking task,the target cannot keep accurate tracking under the interference of scale change,occlusion and other factors.Based on the kernel correlation filter(KCF)tracking algorithm,a long-time target tracking algorithm with anti-occlusion ability is proposed.Orientation gradient histogram feature(HOG)and color feature(CN)are fused to express the target information adequately.The scale filter is constructed to make the algorithm adapt to the target scale change.The average peak correlation energy index(APCE)is introduced to judge the occlusion of the target.After the target is occluded,the SVM classifier is used to re-detect the target position.In terms of model updating,the updating strategy is adjusted by the average peak correlation energy(APCE)value and the maximum correlation response of the position filter.In the OTB100 video set,the improved algorithm was compared with other tracking algorithms,and the experiment showed that the improved algorithm had a great improvement in tracking performance under the disturbance properties such as scale change and occlusion.(2)In view of the deficiency of background aware correlation filtering(BACF)algorithm,an improved background and time aware correlation filtering algorithm based on BACF algorithm framework was proposed.Firstly,considering the invariance of the depth features,the depth features extracted from the deep network are used to replace manual features in BACF to improve the robustness of the tracking algorithm.Secondly,BACF algorithm ignores the consistency of filters in time when updating the filter model.The introduction of time regularization term can prevent the learned filters from bias to the background and improve the anti-occlusion ability of the algorithm.In order to verify the feasibility of the improved algorithm,In order to verify the feasibility of the improved algorithm,experimental verification was carried out on the OTB100 dataset.The results show that the improved BACF algorithm has improved tracking performance in all aspects,and has a better performance than the BACF algorithm in anti-occlusion.