Structure- and Sampling-Adaptive Gait Balance Symmetry Estimation Using Footstep-Induced Structural Floor Vibrations

Abstract

This paper presents a structure- and sampling-adaptive approach for analyzing human footstep-induced structural floor vibrations to estimate footstep ground reaction forces (GRFs) and gait balance symmetry. Balance symmetry and footstep GRFs are critical indicators of overall gait health and elderly fall risks. Prior works, including direct observation by trained medical personnel, computer vision-, pressure sensor-, and wearable-based sensing, are limited due to operational restrictions. We introduce a nonintrusive balance symmetry monitoring approach, which utilizes sparse structural vibration sensing. The intuition is that footstep-induced floor vibration responses are proportional to footstep GRFs, and balance symmetry can be defined using consecutive GRF pairs. However, GRF-vibration relationships are also influenced by spatially-varying structural properties and gait sampling bias, introducing errors to real-world estimations. We address these challenges first by extracting structural regions to overcome spatially-varying vibration behavior and then by developing a kernel-based robust regression model to overcome biased training data and enable robust GRF and balance symmetry modeling. We evaluate our approach through real-world experiments, achieving a balance symmetry index estimation accuracy as high as 96.5%.