Inspection And Matching Technology Of Pedestrian In Multi-horizon Non-overlapping Surveillance Video

Along enlargement of surveillance system, multi-camera wide-horizon surveillancebecomes the mainstream, wherein the camera horizon in most surveillance systems are notoverlapped; its surveillance target is pedestrian; therefore, it is more important to study theissue of target tracking and matching in multi-horizon surveillance system; and it givesmost priority to pedestrian target inspection and matching. Aimed at pedestrian inspectionand matching matter in the surveillance system, it advances a feasible proposal of solvingpedestrian matching in non-overlapped horizon surveillance video based on continuousmulti-frame target information feature fusion through pedestrian inspection, inter-frametarget association and pedestrian matching.This paper firstly introduces the objective and significance of pedestrian targetinspection and matching technology in surveillance video and makes brief introductionabout current research situation at home and abroad.Then, it illustrates several common inspection methods of pedestrian targets and makesexperiment on the surveillance video including pedestrian target and gives the resultthrough analysis; based on the requirements of the matching steps, it selects the pedestriantarget inspection based on CENTRIST feature. Then, it makes introduction andexperiment on several methods about inter-frame target association and selects theassociation of target position and motion information on inter-frame target to obtain targetsequence based on the pedestrian target feature of surveillance video.Finally, it studies the pedestrian target matching technology based on different featuresand matching strategies in details, advances the feature fusion judgment algorithm basedon the advantages and disadvantages of the current algorithm and multi-frame targetinformation, and finally gives the experimental result and analysis on the experimentalresult. The experimental result shows the algorithm advanced by this paper has goodaccuracy and stability.Although this paper carries out certain study on pedestrian inspection and matchingtechnology in surveillance video, the surveillance scene and target are relatively simple.So it needs to have further study on the complex scene and target possibly occurring inactual application.