Research On Human Motion Analysis Based On Vision

Technology of human motion analysis is research of the technology from one or more perspectives on the pedestrian target attitude and behavior, which involves computer graphics, machine learning, pattern recognition, kinesiology and physical optics and other disciplines that can be used in intelligent video surveillance, human-computer interaction, virtual reality, identity recognition. Technology of human motion analysis is mainly to extract human from video sequences, and then extract the features of human target, classify and identify the characteristics of actions.This paper mainly studied the technology of human motion analysis in monocular vision and binocular vision. In monocular vision, the paper used the method of human motion analysis based on two-dimensional shape template matching. The paper mainly put forward an improved thinning algorithm of skeleton extraction that combined with the human prospect of geometric characteristics of rectangle contour and skeleton Hu invariant moments using the K nearest neighbor classification algorithm(K Nearest Neighbor, KNN)to classify and recognize human movement behavior. In binocular vision, the paper proposed a method to detect face and both hands skin color region based on skin color model. Then with the obtained parameters of binocular camera after calibration, the paper calculated the three-dimensional coordinates of the key parts of the face and hands, using the principle of binocular stereo vision. This paper also presented an analysis of human posture recognition method based on vector angle, using the three-dimensional coordinate information of key parts of the face and both hands to analyze human body motion based on the change of attitude and velocity. The main innovations of this paper are as follows:(1)In the monocular vision, this paper proposed an improved Zhang-Suen thinning algorithm, and applied it to human action database to extract moving human skeleton, and calculated the Hu invariant moments of the human skeleton, the rectangle degree of human body foreground, circular degree geometric features such as the human body movement and prospects as the data characteristics of human behavior as the training and testing of K nearest neighbor algorithm, finally distinguished the human action categories.(2) In the binocular vision, this paper put forward a face and hands of the skin region detection method based on YCbCr space. By detecting the face and hands, the paper calculated the value of mass center in the color image region, and proposed a kind of sparse block matching algorithm based on neighborhood color region centroid. The algorithm is mainly used by calculating the value of gray mean and variance of the target area of the two images, and calculates the correlation coefficient about the centroid of neighborhood to find the optimum matching of the centroid; In the process of human action recognition, the paper used binocular stereo vision principle to calculate the3D coordinates of the skin color region centroid of face and hands, and put forward an analysis method based on hand gesture triangle, and developed an estimation system of binocular vision of human motion in depth and velocity.