Research On Human Body Parts Labeling Based On Video Sequences

Over the past decade, the visual analysis of human movement is concerned by many researchers in the field of computer vision. It becomes a new research of computer vision. It can be used in many applications such as smart surveillance, perceptual interface, virtual reality, motion analysis and so on. Its broad application prospect greatly stimulates the people's enthusiasm. So, it has high academic value and theoretical significance, and has become one of the hot issues in recent research.This paper mainly studies the segmentation and location of motion human body parts under the circumstances of single camera image sequences. At first, this paper studies how to labeling the head, how to segment the human body parts and how to identify each part of human body. At last, some significant parameters, which are very useful to the further research of visual analysis of human movement, are obtained.In this paper, the main work and characteristics are as follows:Firstly, Classic head labeling based on the Torso-Mask firstly is given. Aiming at its disadvantages, a new multi-degree of freedom physical model is proposed, and a corresponding algorithm is studied. In which introduces the trunk angle factor, head deviation angle factor, and the neck length parameter to get better model performance. Experimental results show that the proposed multi-degree of freedom model and its corresponding algorithm has good performance in terms of the head positioning accuracy, time-consuming, and the robustness.Secondly, there are large numbers of non-zero pixels statistics in the front-ground of an image. And the square region has an arbitrary rotation angle, so the statistics is slow if we traverse the image to count them. Therefor the line integral image is proposed, and applied to accelerate the head positioning algorithm.Thirdly, classic triangulation based on interval point of human body contour can not adapt to the human body in different postures and the changes of human size in image. A new Constraint Delaunay triangulation based on the body contour corner is proposed. It can realize the extraction of the human skeleton and the partition of each part of human body, so it enhances the robustness of the algorithm. Meanwhile, a new parameter-extraction method for Six-Segment Model is proposed in the process of identifying each part of human limbs. Further, a more effective human body model is studied in two different view, i.e., front view and side view. Finally, sum up the work of this paper. And further, head location in the case of many upright persons is studied, and some simple human behavior analyses are discussed.