Research On Moving Object Tracking Of Visual Target Under Complex Scenario

Moving target tracking is a popular research directions of computer vision technology in recent years.It has been widely used in intelligent surveillance,medical imaging,automatic driving,visual navigation and other fields.However,due to the complexity and variability of tracking scene,while facing fast motion,motion blurring,shape change and other changes in the appearance of the target,as well as noise interference,illumination change,low resolution and other environmental changes,tracking drift is easily caused by using a single feature,and the main reason for tracking failure is that the target is occluded.In this thesis,the difficulties of target tracking in complex scenes are analyzed.On this basis,the traditional correlation filtering algorithm is introduced.Based on the advantages and disadvantages of the correlation filtering algorithm,a framework of target tracking algorithm based on multi-feature fusion and re-detection is proposed.The main contents of this thesis are as follows:(1)The related theory of target tracking is studied,several common features in target tracking are studied,and their performances and characteristics are analyzed.A strategy of using multi-feature to enhance the adaptability of tracking scene is proposed,which makes the tracking algorithm more conducive to tracking in various scenarios.(2)The related theory of correlation filtering method is studied,and the basic theory and steps of the algorithm are introduced.In order to overcome the shortcomings of correlation filter tracker which can no evaluate current tracking effect,a tracking effect evaluation module is added to enable the tracker to judge whether the target has been lost.(3)Aiming at the problem that the target cannot continue tracking after occlusion occurs in tracking,the re-detection module is introduced.According to the output of the tracking effect evaluation module,when the target is lost,the target is re-detected in the image,and the tracking is restarted after the target is found.(4)Finally,this thesis evaluates and compares several mainstream trackers and the trackers proposed in this thesis on large-scale open datasets.Experiments show that compared with the traditional algorithm,the tracking effect of this algorithm in complex scenes has achieved satisfactory results.