Can Markerless Pose Estimation Algorithms Estimate 3D Mass Centre Positions and Velocities during Linear Sprinting...

Area of Science:

• Biomechanics and Sports Science
• Motion Capture Technology
• Human Movement Analysis

Background:

• Accurate, non-invasive measurement of 3D mass centre (MC) and its derivatives is crucial for understanding sports performance demands.
• Markerless human pose estimation offers a potential alternative to traditional marker-based motion capture systems.

Purpose of the Study:

• To evaluate a markerless human pose estimation method combined with Kalman smoothing for measuring mass centre velocities.
• To assess the accuracy and precision of this method across different activities, including sprinting and skeleton push starts.

Main Methods:

• Synchronous capture of marker (Qualysis) and markerless (OpenPose) motion data during sprinting and skeleton push starts.
• Calculation of mass centre positions and velocities from raw markerless pose estimation data.
• Application of Kalman smoothing to refine velocity measurements.

Main Results:

• Raw markerless pose estimation yielded significant errors in mass centre velocities for both sprinting and skeleton pushing.
• Kalman smoothing reduced mean error in horizontal mass centre velocities during sprinting but precision remained limited.
• Markerless methods showed poorer performance and higher errors for activities with unusual body poses, like skeleton pushing.

Conclusions:

• Markerless pose estimation with Kalman smoothing shows potential for non-invasive mass centre velocity measurement, particularly in activities like sprinting.
• Current algorithms exhibit limitations in accuracy and precision, especially for complex or unusual movements.
• Activity-specific model retraining is recommended to improve the generalizability and reliability of markerless pose estimation for diverse sports applications.