Research On Human Abnormal Behavior Detection And Analysis Based On Video

Human behavior analysis based on video is a hot research topic in computer vision field.With the development of China's aging population,video analysis technology has been widely used in the field of intelligent monitoring.Under such an application background,this paper analyzes and studies the behaviors of two special groups,such as the elderly and the patients.This paper studies five kinds of normal and abnormal behaviors for detection and recognition,including walking,chest pain,headache,squat,falls,etc.Firstly,the detection and tracking of moving human in the scene is realized,and then the behaviors are analyzed and judged on the basis of the feature extraction.Finally,the five behaviors are classified and identified.This paper mainly does the following work according to the three parts mentioned above:1)In the moving target detection,although the vibe foreground detection algorithm is simple and fast,but it would not be able to extract complete moving targets.In view of the shortcomings of Vibe algorithm,the improved Vibe foreground detection algorithm is proposed,which combines Vibe with a series of image processing methods,including area threshold method and morphological processing method.A moving object type separation strategy is proposed in this paper.The human body and the nonhuman body region are separated based on the detection of human head,and the research object is further simplified.2)In the part of feature extraction,this paper introduces in detail several single feature extraction methods mainly includes feature extraction based on the shape,focusing on Hu moments and Fourier descriptor feature;based on the geometric characteristics of the human body extraction,using the minimum enclosing rectangle box to get body ratio of height to width,the inclination of the body characteristics;human motion characteristics,including changes in the centroid,the speed etc.In the end,a feature extraction scheme based on multi feature fusion is proposed for using one single feature to distinguish the behavior always makes mistakes.3)In the classification recognition of abnormal behavior,a key frame extraction technology is proposed to reduce the data quantity of video image processing.As huge amount of video processing will lead a very low efficiency.Then the feature extraction is carried out,and the features of the image are studied and the combination features are stored in the feature vector.Finally,the KNN classifier is used to train and classify the samples.The effectiveness and feasibility of the method proposed in this paper are verified by experiments.