Research On People Counting Algorithm Based On Binocular Stereo Vision

People counting is an indispensable decision-making work for urban construction and commercial operation.Aiming at the problem that the current people counting monitoring scene is easy to be blocked by pedestrians and environmental factors,this paper proposes a people counting method based on binocular stereo vision.By adding depth information,it can effectively overcome the influence of pedestrians and environmental changes,and realize people counting in the monitoring area automatically and accurately.The main work and achievements are as follows:(1)A semi-global matching algorithm based on improved Census transform is used to obtain disparity map.This algorithm combines the features of AD(Absolute Difference),GRD(Gradient)and fast Census,and improves the quality of disparity map by fusing pixel strength,gradient boundary and texture structure information.It solves the difficult problem to obtain high matching accuracy in current semi-global stereo matching algorithm with single feature.The results show that compared with the semi-global matching algorithm based on Census transform,the proposed algorithm can effectively reduce the pixel error rate of disparity map,and improve the quality of disparity map to achieve effective dense disparity map acquisition,The acquisition obtains a clear and complete disparity map of the pedestrian's head,and meet the needs of target detection based on depth information.(2)The target detection algorithm based on depth information is designed.The depth layer is used to extract the depth layer of the target information and locate the specific position of the target.The depth information is introduced to provide more additional information for target detection,and an improved elliptical fitting target detection algorithm based on depth information is realized.The results show that the proposed algorithm can retain more information of the target and can achieve accurate target detection.(3)The target tracking algorithm based on multiple features fusion is implemented.The position,shape and height features of the detected target can be obtained more accurately by using depth information.This paper constrains the fluctuation of the target position by adding the coordinates,area and height information of the target.Enhancing the stability of target features to reduce false tracking,and achieve target tracking.Finally,target counting is performed by setting a virtual detection line.The feasibility of the algorithm is proved by experiments,and the statistics of passenger flow is realized.