A real-time falls detection system for elderly

A real-time fall detection system is proposed to distinguish various falls during daily activities. Falls are detected in two steps: first a hierarchical algorithm is used to classify the motion and motionless postures such as lying, sit-tilted, sit-upright, standing and walking; it then analyzes whether the current lying or sit-tilted postures are normal or abnormal, based on posture transition and users´ current position. If an abnormal lying or sit-tilted posture is determined, a fall alert will be delivered immediately; if a possible fall is raised (such as normal lying but on the ground), then a music based alert starts playing, and a fall or normal lying will be determined according to whether the user stops the alert music. The advantages of the approach are that it can distinguish various falls efficiently (in real-time within a smart phone), and can also significantly improve the “true positives” for the slow falls with a sit-tilted posture, as well as the “true negatives” for the normal lying compared to the existed fall detection algorithms.