Improving the performance of fall detection systems through walk recognition

Social problems associated with falls of elderly citizens are becoming increasingly important because of the continuous growth of aging population. Automatic fall detection systems represent a possible answer to some of these problems, as they are useful to obtain help in case of serious injuries and to reduce the long-lie problem. Nevertheless, widespread adoption of these systems is strongly influenced by their usability and trustworthiness, which are at the moment not excellent. In fact, the user is forced to wear the device according to placement and orientation restrictions that depend on the considered fall-recognition technique. Also, the number of false alarms generated is too high to be acceptable in real world scenarios. This paper presents a technique, based on walk recognition, that increases significantly both usability and trustworthiness of a smartphonebased fall detection system. In particular, the proposed technique automatically and dynamically determines the orientation of the device, thus relieving the user from the burden of wearing the device with predefined orientation. Orientation is then used to infer posture and eliminate a large fraction of false alarms (∼ 98%).