The Human Daily Behavior Recognition For The Family Environment

With the deepening of the degree of aging in China, the intelligent home care has been put forward higher requirements. It is to ensure that the full range of people living in a sense of security and quality of life. Normal behavior can be detected and processed, as well as a variety of unexpected behavior. System can be early warning to reduce human losses, while not affecting the daily lives of people. At present, with the improvement of the performance of computer operation and storage, computer vision has been applied in the realization of complex human-computer interaction. It has many advantages in the realization of intelligent home monitoring. Video based behavior detection and recognition is one of the important research topics in the field of computer vision. The research direction combined with the pattern recognition, artificial intelligence, image processing, and other aspects of technology. The technology has many applications in video surveillance, intelligent video analysis, anomaly detection, human-computer interaction and so on. In this paper, the human behavior representation and recognition technology based on video is discussed and studied.In this paper, firstly, the behavior of moving objects in the scene are detected and processed. Background subtraction method based on mixed Gauss model is used to separate the background and moving objects. Then the extraction result is processed by mathematical morphology that is open operation and close operation, and then canny operator is used to extract the contour of the moving object.Secondly, about the extraction of human behavioral characteristics,it includes the relative distance between extreme points and the centroid、velocity about centroid changing、rectangle aspect ratio of the target. And the feature vector serialization of the video image, for hidden Markov model. The three basic problems were based on valuation problems, learning problems, decoding problems, using the Baum Welch algorithm and the forward algorithm for experimental design, in order to achieve the realization of human daily behavior system.This paper uses two databases, Weizmann database and self-built database. In the part of the video as an example, analysis and comparison of multiple human behavior contour feature parameter curve. Finally two databases were used cross validation method to test the effect of the algorithm. Experimental results show that will human behavior contour features and hidden Markov model combination can be constructed with simple and practical human behavior classifier, conducive to the development of a family of real-time uninterrupted intelligent monitoring.