Accessible In-Home Gait Assessment Using Spatiotemporal Neural Networks with Visual and Kinematic Data

Gait analysis traditionally relied on controlled laboratory settings, limiting its practical use in non-clinical environments. This study proposes an accessible framework utilizing consumer electronics, combining data from an Apple Watch and a visual sensor system, to measure stride time (ST) across three walking speeds: slow, normal, and fast. Data was collected from eight participants in a semi-controlled setting designed to match real-world conditions. The machine learning framework, combining Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks, was utilized to analyze the multimodal sensor data and calculate ST. The system demonstrated strong agreement with an infrared marker-based optical motion capture system particularly at slower walking speeds. These findings underscore the feasibility of combining consumer-grade wearable and ambient sensors for accurate, accessible gait analysis in nonclinical settings.Clinical Relevance-This framework offers a cost-effective solution for gait analysis, reducing reliance on expensive clinical equipment. By utilizing consumer electronics, it provides a user-friendly and accessible alternative for individuals with mobility impairments, enabling regular assessments in non-clinical settings.