Lower-Limb Falling Detection System Using Gated Recurrent Neural Networks

Abstract

Accidental falls are one of the most common causes of premature disability and mortality related to unnatural causes. This affects mainly the elderly population. With the current aging of the population, the rate of accidental falls increases. Computer systems for gait analysis and fast assistance in ubiquitous environments can be effective tools to prevent these accidents. In this article we present the advances in the creation of an intelligent device for detecting falls and risk situations based on accelerometer signals registered on the user’s ankle. The proposed method makes use of Deep Learning techniques, specifically Gated Recurrent Neural Networks. The results show that the proposed model is a viable alternative to detect falls and fall risk, which can be implemented in low performance devices for greater autonomy, lower cost and comfortable portability. These results open the possibility of combining fall detection with a biomechanical analysis system to identify gait deficiencies and their relation with falls.