Research On Human Behavior Identification Technology

Moving human body detection and behavior identification in video sequences is a task related to video image processing, computer vision, pattern recognition, artificial intelligence and other areas. It is an important research topic in recent several years because of its widely application in business, medical and military fields. Since human behavior is diverse and non-rigid and video image is very complex, research on a truly robust algorithm of human activity recognition is still a challenging task.In this paper, moving human body detection and behavior identification is researched. For the moving objects detection, Gaussian mixture model is initiated by k-mean clustering, which saves storage space and makes the initial Gaussian mixture model more conform to the background. Because different time and areas in the scene need to the different number of Gaussian model, Gaussian composition of the Gaussian mixture model is chosen adaptively. Redundancy Gaussian components are removed to save storage space and improve the detection speed. When the shadow is detected and removed, in order to overcome the disadvantages of the present shadow detection and remove method, an improved shadow detection method is proposed. It needn't set threshold in advance and only detects moving target or shadow region that is detected by Gaussian mixture model and removes shadow. The accuracy of shadow remove is improved and processing time is decreased.In order to solve the scaling sensitivity of Radon transformation, an improved Radon transformation is used to extract Radon features of minimum enclosing rectangle of motion human for every frame of video sequence. And the height-width ratio of the rectangle is extracted. The improved Radon transform is the invariant to translation, rotation and scale change. Therefore, normalized processing of size is not needed before feature extraction and motion description, which is more robust and benefits to the following human behavior analysis. When human behavior is recognized, a human behavior recognition method based on subsectional two-dimensional principal component analysis (subsectional 2DPCA) is proposed. The robustness of behavior identification is improved. The experimental results show that the proposed method can effectively identify human behavior and has low computation and higher recognition rate.