Locally weighted PCA regression to recover missing markers in human motion data

Area of Science:

• Computer Vision
• Biomechanical Engineering
• Data Science

Background:

• The "missing markers problem" is a persistent challenge in motion capture.
• Existing methods struggle with incomplete marker data during motion capture sessions.

Purpose of the Study:

• To address the challenge of missing markers in motion capture.
• To develop a novel regression method for improving motion tracking accuracy with incomplete data.

Main Methods:

• Proposed a locally weighted principal component analysis (PCA) regression method.
• Introduced sparsity constraints into traditional least squares methods via multivariate tapering.
• Developed a novel least squares method incorporating sparsity constraints.

Main Results:

• The proposed regression method achieves high estimation accuracy.
• Demonstrated good numerical stability in experimental results.
• Successfully addressed the missing markers problem in motion capture.

Conclusions:

• The locally weighted PCA regression offers an effective solution for missing marker data.
• This method provides a robust and accurate approach to motion capture challenges.
• Represents a significant advancement in least squares methods with sparsity constraints.