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

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning

Background:

• Traditional gait analysis is confined to labs, limiting real-world application.
• There is a need for accessible, cost-effective gait monitoring solutions outside clinical settings.

Purpose of the Study:

• To develop and validate an accessible gait analysis framework using consumer electronics.
• To measure stride time (ST) across various walking speeds using multimodal sensor data.

Main Methods:

• Utilized an Apple Watch and a visual sensor system for data collection.
• Employed a machine learning framework combining Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks.
• Collected data from eight participants in a semi-controlled, real-world-simulating environment.

Main Results:

• The proposed system demonstrated strong agreement with infrared marker-based optical motion capture, especially at slower walking speeds.
• Successfully calculated stride time (ST) across slow, normal, and fast walking speeds.
• Validated the feasibility of using consumer-grade sensors for gait analysis.

Conclusions:

• Combining wearable and ambient sensors offers a feasible approach for accurate, accessible gait analysis.
• This framework provides a cost-effective, user-friendly alternative for regular mobility assessments in non-clinical settings.
• Enables individuals with mobility impairments to undergo regular gait assessments conveniently.