Research Of Indoor Fall Detection For Elderly Based On Video

In China,with the growing of population aging increasingly,the problem of empty nesters become more and more prominent.Currently,Fifty-four percent of urban elderly households are empty-nest families.In the world,the fall has become the top killer of the elderly accident.If accidental fall happened at home,it is likely to cause disability without timely treatment,even death.Therefore,it is very important to research and construct the system of fall detection for elderly.Now,Taking an overview of the study about fall detection,most of them are based on the video images fall detection and wearable sensors fall detection,as the elderly have some exclusive on wearable sensors and the sensors to obtain data easily lead to errors,thus fall detection based on sequence of video images get higher practicability.This paper proposed and implemented a video-based fall detection system,aimed at real-time,accurately detect the behaviors of fall and to win precious time for the treatment of the elderly.This system contains four modules,including the human target detection,target tracking,multi-feature extraction and classification detecting.Firstly,this paper applies improved single Gaussian modeling to detect the object of video foreground,introducing the sub-block histogram,we compare with the histograms distance of two adjacent frames from the same area to determine whether to update the background.This method can quickly adapt to light mutation requirements,at the same time,it can reduce the situation of the slow-moving objects to be integrated into the background to some extent.Since the extracted foreground contains some shadow,it uses the shadow detection algorithm that based on adaptive threshold of HSV color space,and can effectively remove the shadow,to get more precise target shape of the human body.And then,the target should be tracked stably in order to adapt to the situation that a plurality of human targets appear in the monitoring scene.In this paper,a new algorithm based on color histogram matching and Kalman filter is used to track the target,this method can effectively deal with the human target appears split,adhesions,at the same time could use Kalman filter prediction matching for the missing target.The system is followed by continuous tracking,,which lays the foundation for the subsequent extraction of the human body.Finally,this paper presents a fall detection algorithm based on multiple feature analysis and SVM classification,using human center of mass,aspect ratio,angle,the ellipse axle ratio,and the vertical histogram by Fourier transform as effective features,and the method based on sliding window is proposed to extracting the continuous human target feature to build a feature vector space,which could be put into the offline training completed SVM classifier to make classification decisions.In this paper,the proposed approach can real-time detect the fall in indoor scene,Experiments prove that the average recognition rate reached 90.95%.This method could be used in homes,hospitals,nursing homes and other fields,for protecting the security of the elderly.