Agreement Between Sagittal Foot and Tibia Angles During Running Derived From an Open-Source Markerless Motion Capture...

Area of Science:

• Biomechanics
• Computer Vision
• Sports Science

Background:

• Markerless motion capture systems offer accessible 2D kinematics tracking using standard video cameras.
• DeepLabCut is an open-source platform for markerless pose estimation.

Purpose of the Study:

• To evaluate the performance and accuracy of the DeepLabCut platform for analyzing lower limb 2D kinematics in runners.
• To compare markerless motion capture with traditional manual digitization methods.

Main Methods:

• Trained a deep neural network using 2D sagittal plane videos of 50 runners' lower legs for foot and tibia pose estimation.
• Validated the trained model on videos from 34 new participants.
• Calculated foot and tibia angles for 7 strides using both markerless and manual digitization, assessing agreement with mean absolute differences and intraclass correlation coefficients.

Main Results:

• The trained DeepLabCut network achieved train/test errors of 2.87/7.79 pixels (0.5/1.2 cm).
• Markerless methods systematically overestimated foot angles and underestimated tibial angles compared to manual digitization (P < .01).
• Excellent agreement was observed between markerless and manual methods, with mean differences ≤1° and intraclass correlation coefficients ≥.90.

Conclusions:

• Open-source markerless motion capture, specifically DeepLabCut, is a reliable and accurate tool for analyzing human lower limb motion.
• This technology presents a promising, accessible alternative for biomechanical analysis in research and clinical settings.