Detection And Tracking Of Pedestrian Target In Video Sequence

In the field of computer vision and image processing,moving target detection and tracking are undoubtedly the key and difficult parts.Currently,the detection methods of moving targets mainly include background difference method,frame difference method and optical flow method.Traditional target tracking algorithms mainly include Hog+SVM tracking algorithm,DPM tracking algorithm,YOLOv1 tracking algorithm,YOLOv2 tracking algorithm and so on.In this research,first of all,the traditional three frame difference method is analyzed,then the method is optimized,namely at first three continuous frames of image are obtained,and then they are processed by bilateral filtering,median filtering,image gray level.After that,the moving target area is extracted,and in the elected moving target area,a minimum area of the rectangular area which may contain the moving target is found.According to the previous step,a small rectangular box rectangle is took out.After that,the closed contour formed by the edge of the rectangle region is extracted and filled with white color to complete the contour.Then the algorithm is applied to the relatively complex scene,and is used to detect the moving target so as to improve the accuracy of the moving target detection.Experimental results demonstrate the effectiveness of the proposed algorithm.In this paper,a double CodeWord structure is designed,and an improved codebook algorithm—DWCodeBook algorithm is designed and implemented by adding real-time update mechanism to the codewords in the codebook in the background modeling process and detection process,which improves the accuracy of target detection.The effectiveness of the algorithm is verified by experiments.In addition,aiming at the fast speed of YOLOv1 algorithm,but the low target detection rate,this paper improves YOLOv1 algorithm.On the one hand,it improves the detection rate by changing the scale of the underlying network several times.On the other hand,it optimizes the level of network structure many times to achieve the purpose of extracting the target area and tracking the accuracy of the target.