Improved real-time gait phase detection using simple force sensors, sequencing conditions and smart fault management

Abstract

Gait event analysis by wearable devices is still a current challenge. Ground foot contacts, lower body segment accelerations or rotations or joint kinematics are commonly used, associated with functional analysis or machine learning algorithms, and currently allow to detect up to 5 different phases during a gait cycle. However, these detection are not always reliable or usable in real time because of calculation duration or sensor faults and drifts. To get more precise and realistic sequencing of the gait cycle, we chose to only use the information of heel or toe contacts with Force Sensing Resistors (FSR) to decompose the cycle in 8 different phases that we determined previously by a bottom-up analysis of the possible contacts during walk. Based on this definition different detection algorithms are presented, which introduces sequencing conditions and smart fault management of sensors information to improve the performance and robustness of the detection. Proposed algorithms are assessed on the recorded walking data of 12 participants and in an online experiment, showing an average of 99% recognition rate of gait phases during the walk. These proposed approaches could offer new possibilities for using simple and low-cost instrumented soles to perform quantitative assessment of walking strategies.