Predicting Missing Marker Trajectories in Human Motion Data Using Marker Intercorrelations.

Area of Science:

• Biomechanics
• Motion Capture Technology
• Data Science

Background:

• Motion capture data often suffers from missing information due to marker occlusion or detachment.
• Reconstructing corrupted marker trajectories is crucial for accurate human gait analysis.

Purpose of the Study:

• To develop and validate a novel algorithm for reconstructing corrupted marker trajectories in human gait datasets.
• To assess the algorithm's performance across various gait patterns, including healthy and pathological gaits.

Main Methods:

• Utilized principal component analysis (PCA) to capture marker inter-correlations.
• Implemented a novel weighting procedure for data-driven reconstruction without requiring training data.
• Tested the algorithm on datasets with simulated gaps in marker data, including healthy subjects and individuals with cerebral palsy.

Main Results:

• Achieved reconstruction errors of ≤ 3 mm for well-suited datasets (healthy gait), even with up to 70% data gaps.
• Median reconstruction errors ranged from 5-6 mm for less-suited datasets (complex gaits).
• Demonstrated the algorithm's ability to handle long gaps and fill data anywhere within the dataset.

Conclusions:

• The proposed algorithm offers a viable alternative to conventional and state-of-the-art gap-filling methods in motion capture.
• The data-driven approach is robust, handling extensive data loss in human gait analysis.
• Limitations include the lack of musculoskeletal constraint enforcement and reduced accuracy with highly unpredictable movement patterns.