Research On Detection And Recognition Of Human Abnormal Behavior Based On Deep Belief Network

Surveillance cameras have been widely used in daily life.Since users usually record the visual output of the camera,relevant personnel can only deal with the results of the record and can not give full play to its role of real-time supervision when an abnormal situation occurs.The monitoring system we need is based on artificial intelligence,intelligent recognition and analysis of the moving targets in the video image frame not only can detect and extract the moving targets in the video,but also can identify the identity of the moving targets or distinguish the moving targets Behavior categories,thereby reducing monitoring costs,improve the real-time video surveillance,reduce unnecessary investment.In this paper,a large number of simulation experiments are done.The training samples and test samples used in the simulation experiments are all mainly based on the Behavior Analysis Database(CASIA)of CAS Institute of Automation.The abnormal behavior of the human body is detected and analyzed through the deep belief network in depth study And classification,mainly on the run,jump,bend over,squat,fall five kinds of human behavior classification identification.The main research content is the analysis and comparison of the abnormal behavior of the human body,including the foreground detection and extraction of the moving target of the human body,the extraction of video key frames and the analysis of abnormal behavior.Based on the analysis and contrast of several commonly used moving object detection algorithms,we choose the mixed Gaussian model(MOG)in the background subtraction method to obtain the static background of the video,and then subtract the background from the foreground to extract the moving human target.Noise and hole for image filtering and image preprocessing,so that the movement of human eyes as complete and clear as possible.In this paper,several common digital image processing algorithms for filtering comparative analysis,the final selection of a new filtering method,based on the control Filtering methods of connectivity domain and morphological processing image the foreground image of the video frame.The paper puts forward the way of extracting the key video frames,screening the surveillance video,selecting the key video frames in the video surveillance through the change rate of the center of gravity and the aspect ratio of the moving human body,intercepting and correlating the key video frames in order to adapt to the depth belief Network(DBN)recognition model;the paper deeply studies the deep belief network in deep learning,and creates a framework of deep belief network that can effectively identify the abnormal behavior of human beings for monitoring video analysis.Finally,through the research on the recognition and analysis of abnormal human behavior,this paper verifies the effectiveness of the algorithm through experiments.On the platform of VS2013,we designed the recognition algorithm using Open CV3.0 visual class library by MFC design system interface to realize the abnormal behavior of the human body Detection system,the proposed algorithm can effectively identify five kinds of abnormal behavior.