The Research Of Recognition Method About Human Behavior Based On Shape Festure

Recognition of human action refers to the analysis and recognition of the motion patterns of human body, it is the hot spot issue in the field of computer vision.and has a broad application prospect in the aspects as human-computer interaction, Intelligent Surveillance, motion analysis and so on. It mainly aims to extract the feature parameters of human behavior, make recognition models and complete human behavior's identification automatically at the base of successful moving target detection. This airticle makes some study on human behavior recognition and analysis which mainly focused on feature extraction and behavior classification.First, the acquisition process of movement human body was introduced briefly. We obtained the binary image sequence of movement body through Background Subtraction, and then did some pretreatment with the binary images obtained.Secondly, this paper descripted at length two types of feature extraction process of human behavior. One is silhouette feature extraction based on Fourier descriptors, obtain the body contour of binary images according to eight neighborhood tracking algorithm, extract the fourier descriptor feature based on center distance of the body contour, meanwhile determine the dimension of fourier descriptors according to the amplitude-frequency graph; the other is the regional feature extraction based on the geometric parameters, via the connectivity processing of the of human body binary image, then we extract the geometric parameters features of the obtained binary image, and finally make PCA with the extracted characteristic parameters to Reduce dimensionality.Lastly, the paper makes close research on the design of HMM classifier and RBF Neural Network Classifier and does some improvement. According to the characteristics of human behavior, it adopts non-cross-type HMM; while during the process of parameter determination of the RBF neural network, it adopts the nearest neighbor clustering algorithm. Through the experiment, the results show that the both recognition methods are effective.