Research On The Elderly Sudden Disability Detection Method Based On Visual Sensor

Sudden disability is a major health hazard for the elderly and a major obstacle to independent living. For protecting the health of the elderly, a system using visual sensor is studied, which can monitor the indoor living status of the elderly who lives alone, timely detect the elderly sudden disability and alarm.By analyzing the main technical route and classic algorithm of the elderly anomaly detection, the problem of disability detection is studied through two aspects of the elderly’s global motions and local small movements. In the global motions, the timeout detection algorithm is designed for detecting the abnormal residence time of the elderly staying in one place. Considering the characteristic of the indoor environment, the visual sensor is installed pointing down vertically. The bounding box selective update strategy is proposed to overcome the failure in obtaining the position of the elderly in indoor space. Surveillance video frame is processed into 4’4 sub region image. The elderly dwell time data statistics algorithms based on regional status marking is designed for counting the elderly’s dwell time in different place. The dwell time data is modeled by using a Gaussian distribution for extracting the elderly’ dwell time threshold in different place to detect the timeout.In the local small movements, through the study of the elderly body motor function, the elderly tiny motion detection algorithm based on the local variance of image and κ-nearest neighbor algorithm is designed. Considering the difficulty of detection tiny motion, the sub region image is divided into 5’5 sub block image for researching the relationship between image local variance and tiny motion. The sub block image variance peak-valley differential rate is extracted. The κ-nearest neighbor algorithm is used for distinguishing the active sub block image and detecting the tiny motion. Then elderly sudden disability detection algorithm is designed by fusing the research achievements in two aspects.The algorithm is realized based on Windows platform with MFC. The key technique of the transplanting of the proposed algorithm to the Android platform is researched. The target detection algorithm is transplanted to Android platform, which provides technical support for the overall algorithm transplantation, and is also a beneficial exploration for the frontend data processing mode.