Fall-off Detection System Based On Head Trajectory And 3D Vision

As the population of seniors is increasing significantly, there are now so many elderly living alone, so the health of the seniors becomes a social focus. Fall-off, as a frequent emergency, may threat the health of elderly greatly. The fall-off detection system can detect fall and offer the elderly first aid, avoiding the situation called long-lie which could deteriorate their health.The fall-off detection system becomes an active research topic in and over aboard. The principal of fall-off detection system is based on the feature of fall-off movement, so that it can distinguish the fall-off and activity of daily life (ADL), and alarm if necessary. Fall-off detection system could assist seniors real time especially who live alone so that risk they face could be reduced.Camera-based fall-off detection system is not worn by users, so less intrusive. By record video, it could offer evidence for analysis the reason of fall-off. The head is sensible during moving and the feature is obvious, so this paper presents fall-off detection system based on head trajectory and 3D vision.3D camera kinect is used in this system. Color and depth image could be read by kinect simultaneously. By background subtraction, the contour of men could be got, so the area nearby head could be located. By skin detection, the skin area could be detected. By both, the head could be located. Based on velocity characteristics of the head, it could distinguish fall-off and ADL.Make record about velocity of head during different types of movement, and analysis feature of velocity by using SVM. First, determine optimal parameter for SVM training by cross validation; second, make study group trained under optimal parameter; finally, predict experimental group by model. The result of this experiment show that this system could distinguish crouch, sit and fall-off, and the sensitivity is over 90%.