Antipodal Video-Based Estimation of Absolute 3D Motion for In-Home Spatio-Temporal Gait Parameter Assessment

Area of Science:

• Biomedical Engineering
• Computer Vision
• Rehabilitation Technology

Background:

• Spatio-temporal gait parameters are vital for gait analysis, with vision-based methods offering low-cost deployment.
• Current vision-based methods struggle with accurate 3D dynamics and robust gait cycle recovery from monocular data, often requiring expert supervision.

Purpose of the Study:

• To develop a self-deployable framework for accurate spatio-temporal gait parameter assessment using dual-view RGB video.
• To enable patient-centric, in-home gait monitoring and long-term tracking.

Main Methods:

• A dual-view 3D keypoint estimation network decoupling absolute 3D keypoint estimation into relative keypoint estimation and reference keypoint trajectory regression.
• Semi-supervised gait cycle segmentation using estimated relative keypoints to enhance robustness and reduce labeling costs.
• Validation on a diverse dataset with variations in camera setups, subjects, and recording sessions.

Main Results:

• The framework achieves an average relative error of 8.54% across 22 standard gait parameters.
• Temporal gait parameters demonstrated high accuracy with only 3.91% relative error.
• The system proved robust across various experimental conditions, indicating reliable performance.

Conclusions:

• The proposed framework offers a practical, patient-deployable solution for accurate, in-home gait parameter assessment.
• This technology facilitates long-term patient monitoring and supports remote rehabilitation efforts.
• The dual-view approach overcomes limitations of monocular systems for precise 3D gait analysis.