Sequential Neural Network-Based Gait Analysis and Step Length Estimation with a Foot-Mounted IMU

Gait analysis with step length estimation is essential not only for assessing an individual's health status, mobility, and balance but also for designing a pedestrian navigation system in infrastructure-less environments. This study presents an improved methodology for gait analysis and step length estimation (SLE) by applying a sequential neural network (SNN), a kind of artificial neural network, to sensor data collected from a foot-mounted inertial measurement unit. Our proposed methodology employs a Butterworth filter and normalization technique to smooth the sensor signals. Key gait-related features are then extracted and fed to the SNN, predicting step length without involving any user-specific parameter. The performance of our proposed SNN-based SLE method is evaluated and compared with our previously developed peak-valley detection-based SLE in terms of accuracy. Ten participants including both male and female (aged 24-50) took part in a walking experiment on a 60-meter linear path, walking under two modes – normal and fast. The accuracy results show that our new SNN-based SLE method achieves superior performance with accuracy exceeding 99.4% for each participant and demonstrates greater resilience to variations in user dynamics compared to our previously developed traditional Peak-Valley detection-based SLE method, confirming the proposed SNN-based approach as a superior solution.