Research And Technology On Human Abnormal Behavior Detection Based On Window-Wise Optical Flow Estimation

Human abnormal behavior detection is an important and interesting field of computer vision research. At the same time, the existing methods mainly include two categories: one method is based on the characteristics of the movement of human for abnormal behavior detection, the main idea of the method is to extract the key features of the movement of the object, such as the speed of movement of the object, the direction of movement of the object, an area, outline, etc. In order to represent the different characteristics of human behavior, although most of these features can be expressed in different human movement behavior under different scenarios, they are susceptible to noise, therefore, it is not very easy to extract these features; the other method is based on human movement trajectory for abnormal behavior detection, the main idea is through the use of rich temporal trajectory information to establish the behavioral model and matching analysis to determine whether there is abnormal behavior, and abnormal behavior should belong to which category, but because such an approach requires real-time stability Human motion detection, segmentation, tracking, or occlusion occurs when the moving target in a complex scenario, the moving object will be to identify less frequently, it is impossible to avoid affecting the detection of abnormal behavioral effects. But both methods need to train a number of sample libraries, with a classifier to classify, so we propose a research and analysis on human abnormal behavior detection based on window-wise optical flow estimation, without extensive sample library to be trained, its main contents and contributions are as follows:First, optical flow method has been widely used for motion estimation of the moving object from video sequences. By assuming that the inter brightness, gradient, smoothness constraints on the basis of the correlation, the current conventional optical flow estimation method of optimization scheme is commonly used by the pixels, which can prone to some errors because the optimization program described using a pixel by pixel local motion objects to solve the problem of displacement is often inadequate. In this paper, we propose a new window-wise by optimization framework to solve the optical flow estimation. Because its ability to describe the motion is for a larger-scale objects and movement, the window optimization can be achieved by a more stable and accurate performance. Through using our method to calculate some challenging cases, experimental results show that the optimization of the light by the window-wise flow is better than the prior by-pixel optical flow constraints.Second, in order to solve the problem about the traditional optical flow in real-time applications, this paper presents the human abnormal behavior detection method based on widow-wise optical flow method. The estimated object motion by the window-wise of optical flow based on weighted moving rectangular area, the direction and the velocity of the optical flow is normalized to determine whether there is an abnormal behavior. Experimental results show that the algorithm can accurately detect abnormal behavior of the human body, with the higher the robustness and low computational complexity to meet real-time requirements.