Research Of Fall Detection Algorithm In Video Based On Mutiple Feature Fusion

Aging problem has become more and more prominent in China,followed by a large number of social problems related to elderly people.On the one hand,the number of anging population is large,on the other hand,the population is aging fast.There are two ways of cares for the aged in China,one is the family supported mode,and the other is the institution supported mode.However,a lot of problems existed in each way of cares.On the one hand,the numer of empety nest elderly has increased fast owing to the influences of aging,more and more elderly are neglected for a long time,traditional family supported mode is no longer fit for the times.On the other hand,the institution supported mode is limited by the shortage of nurses and the heavy labor intensity bothers nurses a lot.With the development of techniques,intelligent elderly guardianship is presented to solve these issues.Fall,as a security issue of concern,has a serious impact on both physical health and mental health of the aged.Rescue time is very crucial for elderly people as it can limit the damage when a fall happenes.That's the reason why more and more experts and scholars devote themselves to researching fall detection system,however,there are a lot of key problems to be solved.The presented fall detection algorithm based on multi-feature fusion belongs to the methods based on videos,also the fall behaviors end of kneeling down and sitting down are addedd in this paper.The main work of this paper includes:(1)Background subtraction with three frames based on median method was presented in this paper,matching with the improved morphological method by retaining the largest connected domain and shadow detection method to obtain the moving object of better quality.(2)A features fusion based fall detection algorithm was presented to describe the changes of shape and velocity,making sure two new fall behaviors could be recognized,the problem of scale change was also solved during this process.(3)Two-layer fall detection system was designed by experiments of self-build fall database,the first layer of recognition was carried out according to the aspect ratios,and the second layer of recognition based on Naive Bayesian classifier was carried out by the fusion of aspect ratios and changes of mass in head area.The recognition rate of fall behaviors is 98.4% and the recognition rate of daily behaviors is 100%.At last,a user-friendly interface was designed by GUI on MATLAB platform.