RecovGait: Occluded Parkinson's Disease Gait Reconstruction Using Unscented Tracking with Gated Initialization...

Area of Science:

• Biomedical Engineering
• Neurology
• Computer Vision

Background:

• Parkinson's disease (PD) is a progressive neurodegenerative disorder impacting motor function.
• Gait analysis is a key tool for early PD detection and monitoring.
• Missing body keypoints in gait analysis, especially from the hips and legs, hinder accurate PD classification.

Purpose of the Study:

• To introduce RecovGait, a novel method for recovering missing human body keypoints in gait analysis.
• To assess the impact of occluded hip and leg keypoints on Parkinsonian gait classification.
• To demonstrate the effectiveness of RecovGait in improving gait analysis accuracy for PD.

Main Methods:

• RecovGait combines a gated initialization technique with unscented tracking to reconstruct missing keypoints.
• Gated initialization provides initial keypoint estimates.
• Unscented tracking refines these estimates to enhance reconstruction accuracy.

Main Results:

• Missing hip and leg keypoints significantly reduce gait classification accuracy for PD from 0.8043 to 0.5217.
• RecovGait successfully recovers occluded keypoints, achieving a Mean Absolute Percentage Error (MAPE) of 0.4082.
• The method demonstrates robustness in mitigating occlusion challenges in real-world gait data.

Conclusions:

• Hip and leg keypoints are critical for accurate Parkinson's disease gait classification.
• RecovGait offers a robust solution for handling missing keypoints in gait analysis.
• This approach enhances the reliability of gait analysis for PD detection and monitoring.