Human Behavior Recognition Based On Template Matching

In the recent years, as computer vision technology improves, the research of human behavior recognition develops rapidly. It has been applied to people's daily life, such as intelligent monitoring, human-computer interaction, virtual reality and motion analysis. It can be said that the application of human behavior recognition in the daily life is linked with everybody closely. It makes us use the way, which is more intelligent and more effective, to instead of the physical labor.Action recognition based on template matching is simpler than other methods. And it is computationally efficient. How to describe the motion and which features to be obtained are significant, which affect the recognition result directly. From these two aspects, this paper does research about human behavior recognition.In this paper, activity recognition is based on successful detection and tracking of the object, so the moving object detection and object tracking are introduced firstly. Then we study some common human behavior recognition methods, such as the method based on Bag of Words, the method based on Hidden Markov Models, the method based on Dynamic Bayesian Network and the method based on Optical Flow Templates On the basis of these, this paper proposes two human behavior recognition methods. The first method is based on feature extraction, which combines the wavelet moments of motion history images and motion energy images and the regional optical flow to describe the features and uses the Euclidean distance to estimate the similarity between the unknown motion and the template. This method is tested on Weizmann database. The second method is based on motion description, which uses the maximal optical flow image and motion history image to describe the motion. This method is tested on KTH database and UT-interaction database. Experiments showed that the both of two methods can improve the recognition result.