Research On Occlusion Problem In Target Tracking Process

In the context of the machine age,all aspects of life have begun to participate in the machine,machine vision is an important part of it.With the extensive application of machine vision,target tracking is one of the most fundamental problems in computer vision,playing an important role in highway transportation,industrial production,public security,military and other fields.The research on tracking algorithms plays an important role in the development and application of machine vision.The significant change in appearance of the target object due to illumination changes,deformation,sudden motion,and severe occlusion during tracking is often a major factor affecting tracking performance.Trackers based on correlation filtering have excellent performance and are extremely robust against challenging conditions that exhibit motion blur and illumination variations.However,the model of learning depends largely on the spatial layout of the objects being tracked,so they are very sensitive to the deformation and occlusion problems.In view of the problem that the tracking accuracy is reduced or even failed due to occlusion during the application process,this paper further studies the target tracking and target detection algorithms based on visible light images.In the occlusion environment,the effective tracking of the tracking target is achieved by an improved algorithm.The main research work of this paper is as follows:(1)This paper proposes a method to judge whether occlusion occurs during the tracking process.When the occlusion occurs,the tracker is stopped and the tracker is kept free of pollution.The accuracy of the tracking is improved to some extent and the drift is prevented during the tracking process.(2)This paper combines the LBP features to improve the real-time tracking of the complementary learning tracker algorithm.Through a lot of experimental research by STAPLE algorithm,it is found that the algorithm can not distinguish the cause of the decrease of tracking accuracy caused by tracking target deformation or occlusion.In this paper,when the tracking accuracy decreases during the tracking process,the LBP feature is used to be insensitive to the target deformation,and the LBP feature is introduced to help determine whether the tracking accuracy caused by the target's own deformation is reduced.When the tracking target is reduced in accuracy due to its own deformation,the tracker remains updated to accommodate and learn the goals of the tracking.When the tracking target is degraded due to occlusion,the tracker does not update the strategy.Experiments show that the improved tracking algorithm improves the tracking accuracy and increases the robustness of the tracking algorithm.(3)In this paper,the fusion target tracking and target detection algorithm is adopted to meet the situation that the target can be detected and the tracking target can be detected when the subsequent target reappears in the case of serious occlusion and long-term occlusion.In order to ensure the real-time performance of the algorithm,the target tracking algorithm is used to track each frame.The target detection algorithm is used as an auxiliary algorithm to activate the detection algorithm to assist the positioning when the tracking target cannot be located.In order to reduce the computational complexity of the algorithm,the tracking algorithm uses the adaptive bilinear interpolation algorithm to reduce the dimension of the target feature.The detection algorithm uses the principal component analysis algorithm to reduce the dimension of the target feature.Experimental verification has greatly improved the speed of the algorithm.