AddBiomechanics Dataset: Capturing the Physics of Human Motion at Scale.

Area of Science:

• Biomechanics
• Human Motion Analysis
• Computational Physiology

Background:

• 3D human pose reconstruction has advanced, but quantifying human motion dynamics (joint torques, forces) remains difficult.
• Existing methods are limited by a lack of comprehensive datasets with high-quality pose and force data for diverse movements.

Purpose of the Study:

• To introduce the AddBiomechanics Dataset 1.0, a novel resource for studying human dynamics.
• To establish a benchmark for estimating human dynamics from motion capture data.
• To provide a publicly available dataset for research in biomechanics and human motion analysis.

Main Methods:

• Collected over 70 hours of motion and force plate data from 273 human subjects.
• Developed novel analytical methods for processing and ensuring the physical accuracy of the human dynamics data.
• Processed more than 24 million frames of data.

Main Results:

• The AddBiomechanics Dataset 1.0 contains physically accurate human dynamics data.
• Novel analytical methods were developed and are reported for dataset construction.
• Baseline results for a proposed benchmark in human dynamics estimation are presented.

Conclusions:

• The AddBiomechanics Dataset 1.0 is a significant resource for advancing the estimation of human dynamics.
• The dataset and benchmark facilitate research into muscle-generated joint torques and external forces.
• Public availability of the data promotes further development in the field.