| With the flourishing development of computer vision technology and the continuous innovation of intelligent security monitoring technology,single cameras cannot meet the requirements of intelligent security monitoring tasks in today’s society due to their limited monitoring ability and small coverage area.Multiple camera groups working together with intelligent collaboration can overcome visual blind spots,improve the efficiency of tracking multiple suspicious pedestrians,and have significant implications for public safety.However,existing tracking algorithms often fail to track multiple pedestrian targets due to factors such as pedestrian occlusion and rapid movement in multi-pedestrian tracking scenarios.In order to achieve continuous and accurate tracking of multiple pedestrians in multi-camera monitoring scenarios,this thesis studies the multi-camera pedestrian target tracking algorithms.The related work is as follows:(1)A summary is given around multi-target recognition detection technology,motion target detection tracking technology,and multi-vision sensing collaborative tracking technology.Then,traditional and intelligent target detection and tracking algorithms and their evaluation metrics are introduced in detail,and some algorithms are compared and analyzed.(2)An overlapping field target swarm intelligence matching algorithm based on multi-vision sensing spatiotemporal relations is proposed.Firstly,an overlapping field determination algorithm based on multiple fusion feature matching is proposed,which accurately generates the overlapping field area and boundary between adjacent cameras based on similarity left gradient,similarity right gradient,and adaptive updating gradient threshold based on spatial difference changes;then,a multivision sensing swarm intelligence matching algorithm based on YOLOv5 is proposed,which enables the main and auxiliary cameras to work together,calculates the distance of pedestrians to the boundary in the overlap view,and further matches pedestrians hierarchically,partitionally,and situationally.Experimental results show that the proposed algorithm can complete pedestrian matching between cameras effectively.(3)A multi-vision sensing target swarm intelligence tracking algorithm based on modified Deep SORT is proposed.Firstly,in order to solve the problem that Deep SORT has poor tracking effect on fast-moving pedestrians,a modified Deep SORT algorithm based on acceleration fitting prediction is proposed,which uses the least squares method to fit the acceleration prediction curve and calculate the average speed,corrects the input speed parameter of the Kalman filter,and reduces the influence of fast pedestrian movement on the Kalman filter prediction;then,in order to solve the problem of tracking failure caused by pedestrian occlusion,a multi-vision sensing continuation tracking algorithm based on secondary matching is proposed,which uses adaptive fusion features and the Hungarian algorithm to perform second-order matching of pedestrians between cameras and corrects the failed pedestrian identity ID of Deep SORT tracking.Experimental results show that the proposed algorithm improves the accuracy of tracking fast-moving and occluded pedestrian targets. |
