Research On Human Target Detection And Tracking Technology Based On Compression Domain

At present,most of the human target detection and tracking algorithms are performed in the pixel domain.Although a small number of researchers have conducted correlation research on the detection and tracking algorithms of human targets in the compressed domain,most of them are aimed at early video compression coding technology,and most of them are based on the human target detection technology in the pixel domain.In view of the existing problems,this paper studies the latest HEVC video coding standard,and proposes a method of human target detection and tracking in the HEVC compressed domain based on motion vectors and coding unit mode information.The research work of this paper is mainly divided into the following three points:Firstly,using data such as motion vectors as human target feature information,a threshold-based human target detection algorithm is proposed.By studying the code stream information of the HEVC compressed domain,in the case of partial decoding,the motion vector information of the HEVC compressed domain is extracted,and by normalization,the motion vector is divided into regions to form a motion vector field,which is then Perform pre-processing operations,and then detect human targets through the time-domain correlation of their motion vector region blocks,thereby detecting human moving targets.Experimental results show that the algorithm can effectively detect human moving targets in the video with the camera fixed relatively.Secondly,taking the motion vector of the HEVC compressed domain as the motion information and feature information of human motion targets,a human target tracking algorithm based on region matching is proposed.The algorithm predicts the human body position of the reference frame of the current frame in the video by forward mapping the motion vector of the current frame,and then uses the histogram of the motion vector direction angle as the matching feature,and combines the region matching strategy to perform the human target in the prediction area.Tracking,so as to achieve the effect of tracking human targets.Experiments show that the algorithm can effectively track human moving targets in videos with relatively still background.Thirdly,aiming at the problem of human flow detection in video,a human multi-target tracking algorithm based on prediction mechanism is proposed.According to the human target area and position information based on the motion vector,the human overlapping area rate between adjacent frames is used to track and count the human moving targets.In order to solve the problem of tracking counting errors caused by occlusion or missing detection,this article will The motion vector combined with the Kalman filter prediction mechanism performs predictive matching on the position information of the human body caused by missed detection,and performs tracking and counting of the human body according to the matching state of the human body in the area.Experiments show that the algorithm has a good tracking effect when the direction of the human body target changes slowly.