The Research On The Detection Of Falls In Indoor Elderly

With the continuous increase of the world's population,the trend of population is becoming more and more obvious,and the number of elderly people living alone is also on the rise.As the same time,the safety of elderly people living alone has become a hot issue of general concern for their children and the society.The fall is an accident that is easy to happen to old people living alone.If they do not find it in time and carry out necessary treatment,in most cases,the delay in time will lead to the occurrence of cardiovascular and cerebrovascular diseases and other diseases in old people,and even threaten their lives.If the elderly can be found in the first time to fall,the elderly take timely and effective assistance,can greatly reduce the injury.In this paper,the method of video signal processing is used to detect the fall of the elderly in the room.In target detection stage,for indoor open to turn off the lights lighting mutation problem,the ViBe of target detection algorithm respectively to improve modeling and model updating stage,then use YOLO model classification and tracking to human body,and extract the body contour height,aspect ratio and fall down three characteristics of area overlap ratio after situation judgment.Different from based on wearable sensors and sensor method based on scene,this method need not old people wear equipment,to prevent any inconvenience due to forget to wear,the method of low cost,equipment loss is extremely low,long life without complex debugging,simple and easy to operate,and is feeling low,for the elderly life habits and the quality have no obvious impacts,easily accepted by the elderly.This paper mainly includes the following aspects:First of all,the existing fall detection methods are introduced,and the fall detection method based on computer vision is further explained.The four steps in the detection process are introduced:moving target detection,moving target classification,moving target tracking and moving target behavior understanding and description.Second,mutations in target detection stage light,and so on and so forth problems,influencing the accuracy of the detection,an improved ViBe of moving target detection method,based on the average brightness change trend of three consecutive frames,whether it meet the characteristics of light mutation,modeling,or to choose to continue to test the next frame,so it can accurately detect moving targets,avoid the interference of external environment due to improve the accuracy of target detection.Finally,in the stage of target classification and tracking,YOLO model is used to classify the targets and track the targets classified as human beings.At the same time,a queue storing the motion features is maintained to determine whether the fall event occurs through the changes of the features in the queue.The experimental results show that this method can accurately classify and track the target,and effectively judge the fall of the target.