Fall Behavior Recognition And Research Based On Multi-features Fusion

Nowadays the society gradually step into aging period,people are paying more and more attention to the elderly people's health.Because of social competition,work pressure and other factors,offspring often can't accompany with their elderly people,which results in a dramatic increase in the number of empty-nest families.In such a social background about global aging and empty-nest families,physical and mental injury pro blems caused by solitary elderly's fall should be solved in great urgent.In order to provide help for the elderly and relieve the injury of the fall timely,researching a set of intelligent behavior monitoring system which can be used in the family or public is very necessary.Given the rapid development of modern intelligent technology and the foundation of summarizing the existing behavior recognition technology,in this paper,we will study a human behavior recognition algorithm based on computer vision entirely.Research work of this paper mainly includes four aspects:First,we build a database of behavior by simulating and recording video about four common behavior in the life of elderly such as walking,sitting down,squatting down and fal ing downSecond,after deeply discussion and research into the common method of target detection: optical flow method,frame differential method and background subtraction method,we proposed an improved target extraction method considering their advantages and disadvantages,this method combine a three-frame differential method and the background subtraction method based on Gaussian background updating model on the basis of weight,and it made up the defects like cavitations of three-frame differential method and sensitivity of dynamic factors of the background subtraction method,which can extract complete target even under the complex environment.Third,on the base of comprehensive analysis of various human behavior characteristic of body position changes,the paper used a multi-feature fusion method for behavior description.Combine body's height,ratio of width to height,the center of mass,the perimeter of circumscribed rectangle with the width of the rate of change,Hu moments and Zernike moments feature together to express the behavior characteristics.Fourth,by comparing a variety of behavior recognition methods,we choose an algorithm of support vector machine(SVM)for classification and identification of human behavior.the paper optimizes support vecto r machine's parameters,then input the fusion features data of the training sample to support vector machine,train classifier and use the trained classifier to predict the behavior to be tested,finally realize the behavior criterion.Behavior recognition experiment was carried out using the above method,the experimental results show that behavior recognition algorithm based on multi-feature fusion has achieved a high recognition rate and can identify fall behavior effectively.If this algorithm can be applied to related electronic equipment,there will be wide prospect of application and great economic value.