Probabilistic inference of multijoint movements, skeletal parameters and marker attachments from diverse motion...

Area of Science:

• Biomechanics
• Motion Analysis
• Robotics

Background:

• Accurate human motion reconstruction is crucial for biomechanical analysis and human-computer interaction.
• Existing methods struggle with inherent uncertainties in motion capture data, including sensor noise and occlusions.

Purpose of the Study:

• To develop a comprehensive probabilistic solution for reconstructing multijoint human movement trajectories.
• To address and quantify uncertainties arising from motion capture data acquisition.

Main Methods:

• Formulated the problem within a probabilistic framework, treating unknown quantities as state variables.
• Employed a Gauss-Newton generalization of the extended Kalman filter for state estimation and system identification.
• Utilized quaternions for spatial rotations and analytically computed sensor residuals and Jacobians for kinematic adaptation.

Main Results:

• Simultaneously estimated joint angles, skeletal parameters, and marker positions with confidence intervals.
• Successfully handled multiple sources of uncertainty, including sensor noise, soft tissue deformation, and missing data.
• Achieved reliable reconstruction of multijoint human movement trajectories.

Conclusions:

• The proposed probabilistic framework provides a robust solution for human motion reconstruction from diverse motion capture data.
• The method offers reliable estimation of kinematic parameters and their uncertainties, suitable for practical applications.
• A software implementation is available, facilitating the adoption of this data analysis tool.