Evaluation of Inertial Sensor Configurations for Wearable Gait Analysis

Abstract

Gait analysis has potential use in various applications, such as health care, clinical rehabilitation, sport training, and pedestrian navigation. This paper addresses the problem of detecting gait events based on inertial sensors and body sensor networks (BSNs). Different methods have been presented for gait detection in the literature. Generally, straightforward rule-based methods involve a set of detection rules and associated thresholds, which are empirically predetermined and relatively brittle; whereas adaptive machine learning-based methods require a time-consuming training process and an amount of well-labeled data. This paper aims to investigate the effect of type, number and location of inertial sensors on gait detection, so as to offer some suggestions for optimal sensor configuration. Target gait events are detected using a hybrid adaptive method that combines a hidden Markov model (HMM) and a neural network (NN). Detection performance is evaluated with multi-subject gait data that are collected using foot-mounted inertial sensors. Experimental results show that angular rate hold the most reliable information for gait recognition during forward walking on level ground.