Research On Detection And Tracking Algorithm Of Moving Object In Complicated Conditions

Moving target detection and tracking has a wide range of applications in the fields of neural networks,machine learning,intelligent transportation,and aerospace.It is a hot research direction in the field of computer vision.However,there are many environmental disturbances in the actual tracking process,such as illumination change,target occlusion,complex background,target deformation and change of camera angle.How to achieve accurate and real-time tracking of moving targets under complex conditions is a major difficulty in the field of computer vision.Therefore,this thesis conducts in-depth research on various moving object detection and tracking algorithms to solve the moving target detection and tracking tasks under various complicated conditions.In this paper,two object detection algorithms based on background modeling and two target detection algorithms based on foreground modeling are studied.The inter-frame difference method and the background difference method based on background modeling cannot accurately establish the background model when the scene has large changes.The optical flow method and HOG algorithm based on foreground modeling can not accurately detect and real-time performance is poor when facing complex situations such as occlusion and illumination changes.In order to realize the moving target detection task under complex conditions,a multi-feature fusion moving target detection algorithm is proposed.Calculating the CLBP histogram similarity SLBP and the hue histogram similarity SHue of the image to be detected,weighted fusion of two histograms to obtain the fusion similarity S between the image to be detected and the background model,separating the foreground target by matching the Gaussian background model and the fusion similarity S of the image to be detected.Experiments show that this algorithm is robust against shadows and complex background interference.Aiming at the moving target tracking task under complex conditions,this paper proposes a particle filter tracking algorithm based on template matching.The template matching technique is integrated into the particle filter algorithm,and the matching of candidate regions and target templates is used to update the particle weights.The target template is updated with the best matching candidate region within a fixed interval of time,and a Levy flight-based cuckoo search algorithm is introduced in the resampling phase to solve the particle degradation problem.On the premise of guaranteeing the tracking accuracy,the number of initial particles is reduced,and the real-time tracking of moving objects is realized.Because of the cumulative errors in the process of target tracking,it is difficult to track the target accurately again when it disappears in the image frame and appears again in the field of vision.Therefore,a long-time single target TLD tracking algorithm is studied in this paper.Aiming at the long-time moving target tracking task under complex conditions,a TLD target tracking optimization algorithm based on Mean Shift is proposed.The initial iteration point of Mean Shift algorithm is determined according to the similarity between the output tracking frame of TLD tracker and the target template,and the final tracking target frame is searched through Mean Shift iteration.Meanwhile,in the tracking module,SUSAN algorithm is used to detect the corners of the target,and SUSAN corners are used to replace the uniform sampling points of the original TLD algorithm,which retains more useful information of the target and reduces the useless points in tracking,thus improving the tracking stability and real-time performance of the original TLD algorithm.