Improved Gait Speed Calculation via Modulation Spectral Analysis of Noisy Accelerometer Data

Chronic diseases among older adults carry a heavy burden on a country’s healthcare system and economy. As such, there is a critical need for the development of cost-effective, technology-based tools that can be scaled to meet the needs of older adults. Gait speed, for example, is an important predictor of change in functional status and health outcomes in older adults. There is no universally accepted method for measuring gait speed in clinical practice and research, and differences in methods may influence the observed associations between gait speed and health. Moreover, existing methods are sensitive to artifacts, which are present in burgeoning low-cost wearable devices. To overcome this limitation, this paper proposes an artifact-robust gait speed calculation method using spectro-temporal signal processing of accelerometer data. To this end, a new so-called modulation domain gait speed (MD-GS) metric is proposed and tested on data collected from forty older adults performing a 400-meter walk test with a sensor placed on a waist-worn belt. Average gait speed calculation is performed for each participant. Experimental results showed the proposed method achieved very high correlation ( $\rho =0.98$ ) with ground truth gait speeds, as well as low errors and error variability (0.05±0.14) m/s, thus substantially outperforming gait speed calculation using a well-known kinematic model. The increased robustness against artifacts, make it a promising solution for aging-in-home applications based on low-cost wearable devices.