JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

How accurately can we estimate spontaneous body kinematics from video recordings? Effect of movement amplitude on OpenPose accuracy

  • 0Neuroscience of Perception and Action Lab, Italian Institute of Technology (IIT), Viale Regina Elena 291, 00161, Rome, Italy. atesh.koul@iit.it.
Behavior Research Methods +

|

January 3, 2025

|

January 3, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

Video-based human body kinematics estimation using OpenPose shows moderate accuracy, performing poorly for small movements but well for large ones (over 10 cm). Accuracy varies by participant and body part.

Area Of Science

  • Biomechanics
  • Computer Vision
  • Human Motion Analysis

Background

  • Accurate human body kinematics estimation is crucial for various fields, including healthcare and industry.
  • Traditional motion capture systems (e.g., Vicon) are accurate but costly and impractical for naturalistic settings.
  • Video-based methods like OpenPose offer a more accessible alternative but their accuracy limitations are not well-defined.

Purpose Of The Study

  • To quantify the accuracy of OpenPose for estimating human body kinematics from plain video recordings.
  • To investigate the relationship between movement amplitude and OpenPose accuracy.
  • To provide guidelines for the effective use of video-based motion capture.

Main Methods

  • Simultaneous recording of 46 participants using OpenPose (video-based) and Vicon (infrared-based) motion capture systems.
  • OpenPose accuracy was evaluated against Vicon ground truth data.
  • Analysis focused on spontaneous movements of varying amplitudes and their impact on estimation accuracy.

Main Results

  • OpenPose demonstrated moderate overall accuracy, with significant variations across participants and body parts.
  • Accuracy was notably lower for small-amplitude movements (< 10 cm).
  • High accuracy was achieved for large-amplitude movements (> 10 cm), with a non-linear (exponential or power) relationship between accuracy and movement amplitude.

Conclusions

  • Video-based human kinematics estimation using OpenPose has limitations, particularly for subtle movements.
  • Movement amplitude is a key determinant of OpenPose accuracy, more so than camera-body distance.
  • These findings offer critical insights for optimizing the application of video-based motion capture in research and industry.

Keywords: