ATHENA: automatically tracking hands expertly with no annotations

Area of Science:

• Biomechanics and Motor Control
• Human-Computer Interaction
• Robotics

Background:

• Marker-based motion capture for studying hand behaviors is costly, time-consuming, and restricts participant movement.
• Existing markerless pose estimation solutions lack validation for precise hand-object manipulation tasks.
• There is a need for accurate, accessible tools for naturalistic hand behavior analysis.

Purpose of the Study:

• To introduce Automatically Tracking Hands Expertly with No Annotations (ATHENA), an open-source Python toolbox for 3-D markerless hand tracking.
• To validate ATHENA's accuracy and reliability against an industry-standard marker-based system (OptiTrack).
• To demonstrate ATHENA's utility in facilitating ecologically valid motor control and learning studies.

Main Methods:

• Developed ATHENA, a Python-based toolbox for markerless 3-D hand tracking.
• Concurrently recorded hand kinematics using ATHENA and OptiTrack systems.
• Compared kinematic variables (grip aperture, wrist velocity, etc.) during unimanual, bimanual, and object manipulation tasks.

Main Results:

• ATHENA demonstrated high spatiotemporal agreement with OptiTrack (R² > 0.90).
• Low root mean square differences were observed for key kinematic variables (<1 cm, <4 cm/s, <5°-10°).
• ATHENA preserved trial-to-trial kinematic variability, yielding identical scientific conclusions to marker-based methods.

Conclusions:

• ATHENA is an accurate, automated, and user-friendly platform for 3-D markerless hand tracking.
• The toolbox significantly reduces financial and time costs associated with motion capture.
• ATHENA enables more ecologically valid studies of naturalistic hand behaviors and human dexterity.