Skeleton-Based Abnormal Gait Detection.

Area of Science:

• Biomechanics
• Computer Vision
• Medical Diagnostics

Background:

• Human gait analysis is vital for diagnosing musculoskeletal disorders.
• Detecting gait anomalies like shuffling or stiff legs is challenging without prior knowledge.
• Existing methods often rely on visual data, limiting their applicability.

Purpose of the Study:

• To develop an automated approach for detecting abnormal human gaits.
• To create a robust normal gait model using skeleton data.
• To distinguish between normal and abnormal walking patterns.

Main Methods:

• A normal gait model was constructed using time-series human joint positions (skeleton data).
• Gait sequences were decomposed into cycles, and postures represented by feature vectors of lower body joint relationships.
• Feature vectors were converted to codewords via clustering to build the normal gait model.

Main Results:

• The proposed method achieved an overall accuracy of 90.12% in distinguishing normal and abnormal gaits.
• Experimental results were validated using both marker-based motion capture and Kinect skeleton data.
• The system automatically determined a normality threshold for anomaly detection.

Conclusions:

• The skeleton-based gait analysis method effectively detects abnormal human gaits.
• This approach offers a promising, data-driven solution for gait anomaly detection in musculoskeletal disorder diagnosis.
• The method demonstrates high accuracy and robustness across different data acquisition techniques.