Tactile-Sensor-Embedded Treadmill for Deep Learning-Based Step-Length Measurement

Abstract

Step-length is a widely used metric for assessing health and disease status. However, existing step-length measurement systems are often expensive or require body-attached sensors, while vision-based alternatives struggle with reliable foot tracking. To address these limitations, we propose a deep-learning-based approach for step-length estimation using a tactile-sensor-embedded treadmill. Baseline step-length was extracted from pressure data using noise filtering and a stacking algorithm. Then, stepwise features were used to refine the estimates through regression modeling, and multiple models were compared using leave-one-out cross-validation. The deep neural network model achieved the best performance, exhibiting a percentage error lower than those of previously reported approaches. These findings suggest that the proposed method offers a cost-effective, nonwearable alternative for accessible quantitative gait analysis.