Research On Cross-Video Pedestrian Detection And Location Technology Based On Object Recognition And Geographical Registration

Cross-video pedestrian tracking is one of the important branches in the field of computer vision.It has a huge development prospect and market value in today's increasingly popular surveillance video system.Existing systems achieve cross-video target tracking mainly by extracting target feature parameters and using video retrieval method.Therefore,the research focus of relevant scholars at home and abroad is to find optimized feature parameters and retrieval methods to achieve better tracking results.Since this method does not solve the problem of describing the geographic location information of pedestrians,it is impossible to obtain the trajectory of the target pedestrian.This thesis mainly studies the geo-registration technology based on object recognition and cross-video pedestrian tracking related technologies.Through application demonstration,a pedestrian trajectory simulation system is established to realize the geographic information positioning of pedestrians.Aiming at the problem of lack of geographic location information in surveillance video systems,this thesis proposes a method of video geographic registration based on object recognition to obtain the geographic location information in video surveillance areas.In view the shortcomings of large amount of computation,high complexity,and weak application ability of the real-time video stream multi-video pedestrian tracking method,a fast head recognition method based on moving object detection optimization is adopted to increase pedestrian positioning speed and obtain pedestrian geographic location information.In the surveillance scenario of multi-video linkage,pedestrian geographic location information as a priori data can be applied to cross-video pedestrian tracking.The main work of this thesis includes the following:(1)A video geographic registration method based on object recognition and spatial feature topology matching.In the aspect of object recognition,a variety of open network training sets combined with 3D model pictures are used to optimize the training set.Compute the control points of the surveillance video imaging plane in conjunction with the ground object spatial database to achieve video geographic information registration.(2)A method for video personnel identification and positioning and cross-video pedestrian detection and positioning: Aiming at the characteristics of large data volume of the multi-video linkage system,a pedestrian detection and positioning method that meets real-time requirements is designed.Combined with pedestrian geographic location information,get pedestrian positioning trajectories across videos.(3)Developed a cross-video pedestrian detection system based on the fusion of video streams and three-dimensional geographic scenes: System includes video access,video display,monitoring playback and scene roaming and other functions.Functions and modules corresponding to relevant algorithms and technologies have been developed to demonstrate the effect of related methods,and to verify pedestrian positioning methods and pedestrian detection methods.