Research Of Human Behavior Recognition Based On Video

Human behavior recognition technology is a hotspot and frontier in the image processing and pattern recognition, and has broad prospects in intelligent surveillance and advanced human-computer interaction and so on, which become one of the most important research in the field of machine vision. Besides, it is the key technologyThis paper deeply analyses the research status and development trend of human behavior recognition technology, and for background subtraction based on the traditional Gaussian mixture background model can’t extract the short stop human target in the scene, many important improvements are applied to the classical Gaussian mixture background model, obtaining better detection effect. Thus the improved algorithm boosts the robustness of the robustness of the system. The main contents are as follows:1) This paper studied the problem that the stationary or short stay foreground object in the scene easily make it difficult to extract the human target of prospect, thus the improved algorithm is proposed based on the traditi onal Gaussian mixture background model, which respectively to improve the update rate and variance in the model, and update online the learning rate of parameter. The experimental results show that the algorithm can accurately extract the background model to improve the human target detection reliability.2) Based on the extraction of human target, this article, according to the action characteristics of human target, selects the shape feature, spatio-temporal interest point feature and Hu moment invariant feature to describe feature vector, and introduce the discrete K-L transform to achieve decision level fusion of spatial-temporal feature and local feature, and draw a conclusion that the fused recognition effect is obviously better than single feature one.3) Developed a human behavior recognition method that combine the hierarchical classification theory with machine lear ning, firstly separate the special actions according to the external shape features of human target, then use the multi-features fusion behavior description operator for behavior modeling, and finally establish learning model. Finally, this article designed a human action recognition system.