MCTN: A Multi-Channel Temporal Network for Wearable Fall Prediction

Abstract

A key challenge in wearable sensor-based fall prediction is the fact that a fall event can often be performed in several different ways, with each consisting of its own configuration of poses and their spatio-temporal dependencies. Furthermore, to enable fall prevention of a person from imminent falls, precise predictions need to be achieved as far in advance as possible. This leads us to define a multi-channel temporal network, which explicitly characterizes the spatio-temporal relationships within a sensor channel as well as the interrelationships among channels by a combination representation of positional embedding and channel embedding to manage these unique fine-grained configurations among channels of a particular fall event. In addition, a transformer encoder is devised to exchange both inner-channel and inter-channel information in the encoder structure, and as a result, all local spatio-temporal dependencies are globally consistent. Empirical evaluations on two benchmark datasets and one in-house dataset suggest our model significantly outperforms the state-of-the-art methods. Our code is available at: https://github.com/passenger-820/MCTN .