The development and evaluation of a fully automated markerless motion capture workflow

Area of Science:

• Biomechanics
• Motion Capture Technology
• Deep Learning Applications

Background:

• Marker-based motion capture is the gold standard but is time-consuming and labor-intensive.
• Markerless motion capture offers potential for more efficient and ecologically valid data collection.
• Deep learning advancements enable automated markerless motion analysis.

Purpose of the Study:

• To present and evaluate a fully automated, deep learning-based markerless motion capture workflow.
• To compare the performance of the markerless system against marker-based motion capture.
• To assess the accuracy of lower limb kinematic data derived from the markerless system.

Main Methods:

• Developed a markerless motion capture workflow utilizing 2D pose estimation and 3D fusion.
• Collected multi-view, high-speed (200 Hz) image data concurrently with marker-based motion capture.
• Computed lower limb kinematic data for 15 participants using the markerless system and OpenSim inverse kinematics.

Main Results:

• High agreement was found between markerless and marker-based motion capture for lower limb joint angles.
• Mean differences for hip, knee, and ankle joint rotations were within the uncertainties of marker-based systems.
• The markerless system demonstrated accuracy suitable for various biomechanics applications.

Conclusions:

• The presented markerless motion capture workflow is a reliable and accurate alternative to traditional methods.
• The open-source, modular nature facilitates integration and promotes transparent development in the field.
• This technology democratizes motion capture, enabling high-quality data collection in diverse settings.