Research On Correlation Filtering Object Tracking Algorithm Based On Image Sequence

The field of computer vision is developing rapidly,and the target tracking of images is a popular research direction in the field of computer vision.Its research content is in image video,which can accurately determine the location information of the target without identifying what the target is.At present,it has been widely used in the fields of defense,intelligent video surveillance,unmanned driving,human-computer interaction and other fields.In recent years,target tracking algorithms have developed rapidly and have a wide variety,but there are also many challenges and problems that need to be solved.For example,in the process of target tracking,they are often subject to changes in the scale of the target,changes in shape,rapid movement,or the target is blocked Interference from internal and external factors.In view of the erroneous results caused by these interference factors in the tracking process,this paper selects the relevant filtering tracking algorithm as the research object to solve such problems in a wide variety of target tracking algorithms.The main work and innovations of this article are as follows:(1)This paper studies the principles of traditional correlation filtering tracking algorithms and the ideas,advantages,and unresolved problems of tracking algorithms related to correlation filtering.And select a variety of excellent algorithms in the two types of tracking algorithms,such as correlation filtering and deep learning,and introduce them in chronological order to solve common problems;draw conclusions through experimental analysis of recurring algorithms.(2)In order to solve the problem that the moving target cannot be effectively tracked under the interference of occlusion and size change during the tracking process,a feature fusion scale name color name tracking algorithm is proposed.The algorithm first uses adaptive fusion of CN color features and LBP texture features,and uses the highest output response after feature fusion as the predicted target position information;then uses the scale filter to make the optimal position information for the predicted target information;and finally Combined with the peak sidelobe ratio of the response graph,the target model is updated and adjusted.The improved algorithm is compared with other tracking algorithms in the OTB-2013 video set.The test results show that the improved algorithm is superior to most algorithms in terms of scale adaptability and occlusion.(3)Aiming at the problem that most target tracking algorithms cannot effectively track due to interference attributes such as target occlusion and deformation during long-term tracking,a long-term target tracking algorithm(CFPE)based on correlation filtering-particle filtering collaboration is proposed.First,the target feature expression in the correlation filter uses the fusion CN feature and the HOG feature to enhance the target description ability in complex situations;then,the average peak correlation energy(APCE)index and the maximum response value are used to make a judgment on the currently detected target position,And decide whether to update the filter template or re-detect the target position;Finally,in the OTB100 and UAV123 video collection and the recent excellent tracking algorithm for comparison test,the test results show that CFPE has stable long-term tracking and real-time,and in deformation and The occlusion is superior to most algorithms.