Research On The Method Of People Counting Based On Video Analysis In Complex Scenes

With the development of video image processing technology,monitoring video has been applied widely in social life.The method of people counting based on video has become an important research direction in video image processing technology.It can be applied to many occasions,for example,the customer number statistics at the inlet and outlet of large stadiums or large shopping malls,the tourist number statistics in scenic spots,attendance assessment of students in Colleges and universities,etc.In this thesis,monitoring video with single fixed monitoring camera is used as a research object.Owing to ubiquitous leakage and misjudgment caused by some factors such as occlusion and similarity interference in the methods of people-counting in various complicated scenes,this thesis puts forward a people-counting method based on combining multi-feature fusion with multitarget tracking in complex scenes.First,video segmentation of the experimental scene is used to extract the key frames of the experimental video segments.Taking the key frame images as the processing object,the physical features,such as hair color,facial skin color and head contour,are extracted respectively.The human head detection method of multifeature fusion is utilized to count the initial number of people,which can exclude error detection caused by part of unmanned areas.Improved adaptive background update target detection algorithm based on five-frame difference is proposed,which can not only exclude unmanned areas in the image,but also solve the problem of discontinuous moving targets and appearing holes easily in the image.Aiming at the detected moving targets,multitarget tracking algorithm is adopted.In different conditions,different association matrices are established to judge the occlusion in complex scenes.The incomplete occlusion of the two heads is tracked by the mean shift algorithm of color histogram,and the rest of the cases are followed by the method of combined heads as a whole,and then the trajectory correlation is carried out.Experiments have been done to prove the feasibility of the algorithm flow.The experimental results show that multi-feature fusing detecting can locate most of the non-occlusion heads.The combination of an improved adaptive background update target detection algorithm based on five-frame difference and the multitarget tracking method can judge the situation of target merging,splitting,disappearance and addition well,and reducing the people statistical error effectively caused by the factors,such as occlusion and interference between targets.