A Smartphone Application for Fall Detection Using Accelerometer and ConvLSTM Network

A fall is defined as an unexpected change in the disposition of the human body, causing it hit brutally the ground. Falls often occur because of external factors that escapes the person's attention. A fall can happen to anyone at any time, however the elderly are particularly affected by these incidents. It can cause simple damages as well as more serious ones, and can even lead to death. Thus, requiring emergency interventions to provide medical assistance. This paper aims to study the problem of automatic fall detection using a phone accelerometer sensor and deep neural networks. We propose a new ConvLSTM neural network architecture for the classification of activities as fall and non-fall. We evaluate the proposed network on two public activities databases and compare with a state of the art network bases on LSTM layers. Moreover, we design and implement a mobile fall detection application.