Measuring Physical Plausibility of 3D Human Poses Using Physics Simulation.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Robotics

Background:

• 3D human pose estimation is crucial for human-environment interaction analysis in AR and action assessment.
• Current metrics focus on joint accuracy but neglect physical plausibility and dynamic stability.
• Existing methods for evaluating pose plausibility assess errors independently, not capturing overall balance during motion.

Purpose of the Study:

• To propose novel metrics for assessing the physical plausibility and stability of 3D human poses using physics simulation.
• To evaluate the effectiveness of physics-based metrics in capturing human pose stability during motion.
• To correlate new physics-based metrics with existing plausibility measures.

Main Methods:

• Developed two new metrics for physical plausibility and stability assessment within a physics simulation environment.
• Applied physics simulation to evaluate predicted 3D human poses from various estimation models.
• Correlated simulation-based stability measurements with existing independent error metrics.

Main Results:

• Physics simulation effectively captures pose plausibility and stability during motion.
• Introduced metrics show strong correlations with existing plausibility measures.
• Demonstrated the utility of physics simulation for evaluating 3D Human Pose Estimation (3D HPE) models.

Conclusions:

• Physics simulation offers a robust approach to evaluate the physical realism of 3D human poses.
• The proposed metrics provide a more comprehensive assessment of pose quality, including dynamic stability.
• This work advances the evaluation of 3D HPE models by incorporating physical principles.