Study Of Recognition Algorithm Of Human Abnormal Postures In Home Surveillance System

With the step of entering the aging society,more and more old people in our country live alone in their home.Due to the aging physiological function,elderly people often fainted caused by some sudden illness,if they can not get immediate treatment,there stands a good chance to threat to their lives.Therefore,the care of “empty nest elderly” is widely concerned in society.With the development of computer vision,home video surveillance system gradually come into people's vision as a new means of security.It can be a good solution to the problem of empty nest elderly care,while saving a lot of manpower and material resources,and solving the worries for many families.Human motion detection in home video surveillance system mainly refers to detecting and tracking of human motion in certain means,extracting motion parameters to represent human motion information,and using these parameters to analysis people's behavior.in this way we can better understand human movement and apply it to practice.The definition of abnormal behavior in this paper refers to the situation that people suddenly fall in normal activities.In this paper,based on studying of the relevant algorithms and theories in this area,we have applied the theories to the detection of human abnormal behavior in home video surveillance system,and have achieved good results.The research contents of this paper are as follows:1.Moving human detection and segmentation As for the problem of slow detection speed and low recognition accuracy in the traditional moving object recognition algorithms,in this paper used Vi BE algorithm to extract the motion foreground image,and as for the shadow problem caused by light in Vi BE algorithm,in this paper,we made improvements: establish double model when initializing the background models,based on the moving foreground detected by gray model,used HSV model to detect shadow and devise it from the foreground image.2.Human features extraction Because the movement of human motion is non rigid motion,single feature can not express human motion accurately,in this paper extracted fusion features from the moving human image,the fusion features include three types of features:(1)Human sketch features,include the ratio of human body height to width,the variation ratio of central point and the effective area ratio,these three parameters are based on the smallest external rectangle extracted from the human body contour.(2)Global features.Extracted Hu moments features to describe the global features of the moving human.(3)Local features.Extracted SURF features to describe the local features of the moving human.Finally,combine these three types of features to a fusion features,based on the experiments,it is proved that the fusion features can roundly describe human motion,the detection accuracy can reach about 98%,and the calculation efficiency is very high,which can well adapt the home video surveillance occasions.3.Classification of human abnormal behavior Because of the high dimension of the feature data extracted in a single frame,using traditional classification algorithm will lead to low efficiency of detection.To solve this problem,this paper used the support vector machine to distinguish normal behavior and abnormal behavior.Support vector machine uses the method of mapping high-dimensional data into high dimensional space to find the optimal classification equation,which reduces the time complexity of high dimensional data classification.The experiments show that the classification method based on support vector machine can effectively distinguish abnormal behavior and normal behavior,and the efficiency and accuracy can meet the practical application requirements.