Research On Surveillance Video Pedestrian Detection And Tracking Based On Particle Filter Algorithm

The development of intelligent surveillance system has enabled more application scenarios for traditional surveillance equipment.In intelligent surveillance systems,pedestrian detection and tracking are widely used in traffic flow statistics,public space security and data mining analysis.As a basis for subsequent analysis and statistical work,it is especially important to improve the accuracy of pedestrian detection and tracking.However,in surveillance video,complex detection and tracking environments place high demands on the algorithm.Factors such as scale changes,morphological changes,background color interference,and illumination mutation reduce the accuracy of pedestrian detection and tracking methods.Therefore,this thesis summarizes and analyzes pedestrian detection and tracking methods under surveillance video,and proposes relevant improvement methods in pedestrian detection and pedestrian tracking respectively.The specific work is as follows:1.The pedestrian detection and tracking methods in surveillance video scenes are studied and compared respectively,and the advantages and disadvantages of each algorithm in surveillance video scenes are analyzed.According to the pedestrian tracking method,the tracking effects of each algorithm are verified on multiple datasets.The improved particle filter tracking algorithm is studied from three aspects:feature fusion,algorithmic theory fusion and adaptive particle filter.2.Aiming at the "ghost" problem of the classic ViBe(Visual Background extractor,ViBe)algorithm when there is a pedestrian target in the first frame,a new improved algorithm-MViBe algorithm is proposed.The specific position of the pedestrian is obtained by the frame difference method,and the background model is reconstructed by multi-frame images.The results show that the MViBe algorithm not only has higher real-time performance,but also can effectively eliminate "ghost" appearing in the classic ViBe algorithm.The algorithm's Precision,Recall,and F-Measure increased by at least 22%,6.6%and 15.8%respectively.3.An adaptive particle filter tracking algorithm GCPF(GLCM-Color PF,GCPF)combining color feature and texture feature is proposed.The color feature and texture feature of pedestrian are described by color histogram and gray-level co-occurrence matrix respectively.The weights of the two features are adaptively adjusted by the contrast parameter of the gray level co-occurrence matrix.The experimental results show that in the complex environment,The GCPF algorithm effectively improves the success rate of tracking when the pedestrian and background colors are similar,and improves the accuracy of pedestrian tracking under the illumination mutation conditions.In summary,this thesis proposes an improved pedestrian detection algorithm—MViBe algorithm in the pedestrian detection part.In the pedestrian tracking part,In the pedestrian tracking part,an adaptive particle filter tracking algorithm based on color feature and texture feature—GCPF algorithm is proposed.The proposed algorithm is validated in multiple sets of surveillance video datasets,experimental results show that the proposed MViBe algorithm can effectively remove the "ghost" in the classical ViBe algorithm,The GCPF algorithm can significantly improve the accuracy and success rate of pedestrian tracking in complex scenes.