Abstract:  The low efficiency of video surveillance in identifying emergencies results in that the recognition system is unable to detect and respond to emergencies in a timely manner, increasing the risk of potential hazards. Therefore, a recognition algorithms of monitoring image abnormal behavior based on computer vision is proposed. Based on the initial background of the monitoring image, a differential operation is used to obtain the differential image between the background image and the monitoring image, and the background subtraction method is used to perform binary processing on the combined sorted new monitoring image to complete target area recognition. Then, a rectangle is used to traverse the target area, collect effective motion blocks from the target area, extract the feature vectors of the motion blocks, and complete the extraction of abnormal behavior features in the monitoring image. And the identification of abnormal behavior in monitoring images through Kuhntak conditions is completed. The experimental results show that the proposed method has an abnormal behavior recognition time of less than 1. 0 s, and the recognition accuracy remains above 94%. It can accurately identify abnormal behavior in monitoring images, effectively improving recognition efficiency and recognition rate.

Key words:  , computer vision, background model, feature extraction, radial basis kernel function, least squares support vector machine(LSSVM) model