A Footwear-Based Methodology for Fall Detection

Abstract

Automatic fall detection is an active research area since several years. Basically, this is motivated by the impact that falls have, in terms of mortality, morbidity, and social costs, which make them comparable to road traffic injuries. The early detection of a fall can be critical to reduce the mortality rate and to limit the associated health consequences. Technological solutions designed to automatically detect and notify a fall may be classified into wearable and non-wearable. Among the former ones, the use of specific devices to be worn by the subject is a very common assumption, but it fails to address user's acceptability issues. In fact, the position of the sensor or its visibility may be perceived as a stigma associated with the primary function of fall detection. To address such an issue, this paper presents a methodology for fall detection that relies on a pair of smart shoes, equipped with force sensors and a tri-axial accelerometer, able to detect a fall and notify it to a supervising system. The instrumented footwear enables the analysis of the subject's motion and foot orientation, recognizing abnormal configurations. The developed algorithm is not computationally intensive, and therefore, can be easily executed on board the wearable device. Laboratory tests provided satisfactory performances in falls detection and correct classification: on 544 falls and 136 activities of daily living, performed by 17 healthy subjects, a 97.1% accuracy has been achieved. Further experiments involving two elderly users demonstrate the effectiveness of the proposed method in a real-life scenario.