Structure-Agnostic Gait Cycle Segmentation for In-Home Gait Health Monitoring Through Footstep-Induced Structural Vibrations

Abstract

This chapter aims to characterize the structural vibrations induced by footsteps to segment a sequence of gait patterns into critical gait phases (including stance phase and swing phase) for in-home gait health monitoring across various floor structures. Gait cycle segmentation is an essential step in quantitative gait assessments for early diagnosis and progressive tracking of neuroskeletal and neuromuscular disorders. Especially, in-home monitoring of peoples’ gait health is beneficial for low-income families and those who have limited access to medical services. Existing studies have adopted cameras, wearable devices, and force plates/pressure mats to segment gait cycles, but they have operational requirements such as direct line-of-sight, carrying devices, and dense deployment, which are not practical for continuous monitoring at an individual’s home. In this chapter, we develop a gait cycle segmentation framework through footstep-induced structural vibrations. The primary research challenges are the complex interplay of the: (1) gait phases and (2) structural properties with the vibration signals. First, gait involves a continuous sequence of multiple types of motions, making it challenging to separate them. Second, people’s living spaces have distinct types of floor structures, leading to difficulty of adapting our framework to multiple structure types. To address the first challenge, we leverage the main insight that human motions at the onset of each gait phase (e.g., heel strike and toe-off) involve unique types of excitation force (e.g., impulsive vs. friction forces). These forces incur peaks at distinct frequency ranges in the responses of the structure. Therefore, we separate gait phases by analyzing the structural responses over various frequency ranges. Second, to make our framework structure-agnostic, we formulate the structural influence on the vibration signals and extract structure-dependent features to represent such influence. Overall, our framework first identifies the structure-dependent dominant frequency ranges for each structure through a time–frequency-domain analysis and extracts vibration signals within these frequency ranges. It then detects time-domain peaks within each structure-dependent frequency range to identify the onset of gait phases. We evaluate our method on two different structures in a real-world setting and achieved consistent results with only a 5% average error in detecting various gait phases.