Research On Multi-pedestrian Tracking Based On Detection Of Vedio

Multi-object tracking(MOT)is a frontier topic in the field of visual surveillance,and its extraordinary military and civilian value has attracted widespread attention.Based on the public monitoring of video and detection,this paper conducts research on MOT technology with the object of moving pedestrian in the video.The main contents of the paper are as follows:(1)Describes the research background and application requirements of MOT,and understand the research and development prospects of offline tracking and online tracking.Focus on the MOT benchmark used in the experiment,including public data sets,annotation information and performance indicators.(2)The offline tracking framework based on network flow algorithm and the generalized maximum multi-clique problem algorithm is studied.According to the characteristics of offline hierarchical association,the network flow algorithm is used as the first layer construction scheme,and the generalized maximum multi-clique problem algorithm is selected as the sub-layer construction scheme.The number of batches of the first and second layers is determined experimentally.(3)The tracking strategy of the object life cycle is compared with the Markov decision process.Combining aspect ratio information to judge the prediction frame,and the kernel correlation filtering is used to secondary predict of the position of the moving objects in frame.The management model of the integrated future frame information is adopted when judging whether the trajectory is terminated.(4)An online multi-object tracking strategy based on relative motion information between objects is studied.In the object matching operation,two conditions of color histogram similarity and bounding box size similarity are added on the basis of position similarity discrimination;The loss objects are recovered by the computational loss method under the constraint of motion structure.(5)The above-mentioned algorithm is used to verify the international MOT benchmark,which indicates its state-of-the-art performance on the public dataset,which alleviates the object occlusion and trajectory fragmentation problems to some extent.In summary,this paper has conducted research on offline and online multi-pedestrian tracking technology based on detection,and verified it on the public international MOT benchmark,and completed the tracking of moving pedestrians under the video.