Influence of the Camera Viewing Angle on OpenPose Validity in Motion Analysis.

Area of Science:

• Biomechanics
• Computer Vision
• Human Motion Analysis

Background:

• Human pose estimation (HPE) is increasingly utilized in biomechanics.
• Limited research exists on HPE algorithm applicability outside controlled laboratory environments.
• Investigating the impact of non-standard camera setups on HPE validity is crucial.

Purpose of the Study:

• To quantify the effect of varying camera viewing angles on human pose estimation accuracy.
• To assess the validity of joint angle estimations using OpenPose compared to a gold standard.
• To determine the optimal camera setup for reliable HPE in biomechanical analysis.

Main Methods:

• Video data were captured from four distinct camera angles.
• Keypoints were extracted using the OpenPose software.
• Estimated kinematic data were compared against a marker-based motion capture gold standard system.
• Accuracy was evaluated for knee, hip, elbow, and shoulder joint angles.

Main Results:

• OpenPose demonstrated reasonable correlations with the gold standard for most joint angles, excluding the shoulder.
• Significant biases were observed in joint angle estimations across different camera viewing angles.
• Back-viewing camera angles generally provided the best performance with the lowest percentage deviations.

Conclusions:

• Camera viewing angle significantly impacts the accuracy of human pose estimation in biomechanics.
• Back-viewing perspectives appear most reliable for joint angle estimation.
• Thorough validation of specific camera setups is essential for diverse real-world applications of HPE.