Research On Human Behavior Recognition Based On High-order Hidden Markov Model

Human behavior recognition has always been a hot research direction in the field of computer vision, its fundamental goal is to make the computer system can like the human eye to recognize and identify the visual information which is collected by all kinds of behavior. In order to achieve recognition effect, we need to comprehensive artificial intelligence, pattern recognition and computer vision technology and then apply them to the recognition of human behavior. With the rapid development of computer technology, people have made big progress in the field of human behavior recognition and some research has been achieved. But the current research can't meet the human behavior recognition technology's requirement which is applied in the practical life, some key problems need to be solved.This paper mainly studied the following aspects of content:(1) According to the conclusion and compared with diffenrent image preprocessing techniques, this dissertation has chosen the adaptive median filtering method and completed the simulation experiment. Through the method of adaptive filtering and histogram equalization method, we have obtained a very good processing result and laid the foundation for the next work. This method can keep image's detail effectively, at the same time, can restrain noise interference. Experiments proved that this method is good to remove noise function.(2) Several typical detection technologies are compared, this dissertation studied the following several kinds of algorithms: optical flow detection technology, background difference detection technology, detection technology of frame difference. According to the actual situation in this paper, this dissertation proposed three frame difference algorithm combining with background subtraction division do foreground extraction and had achieved the expected results through simulation.After a great deal of experimental contrast, on the basis of summarizing the advantages and disadvantages of those methods, this dissertation proposed a new detecting technology-- the fusion of three frame difference and edge detection of moving target detection.(3) Research on feature extraction method. In this dissertation, Haar- like feature extraction method, the shape context feature extraction method and Sift feature extraction method were studied. According to compare Fourier feature extraction method with Sift feature extraction method. This dissertation puts forward a kind of method which is based on the center distance Fourier described human behavior.(4) In this dissertation, the valuation issue of higher-order hidden markov models, decoding and learning problems are studied and applied to the recognition of human behavior. At present, the main is to use the first-order hidden markov model to complete the modeling.,but the assumption that the state transition probability and symbol probability depends only on the current state of the moment. First-order hidden markov model has certain limited human behavior information, reduced accuracy of recognition This thesis puts forward a kind of human behavior recognition based on high-order hidden markov model method. It overcomes the first-order hidden markov model's information incomplete defects, is better to detect the human behavior. this model compared with the first-order model has better temporal structure. At the same time, higher-order hidden markov models can be incorporated into more statistical characteristics. By experiment, the practical application of this model has better functions.