3D Pose Detection of Closely Interactive Humans Using Multi-View Cameras.

Area of Science:

• Computer Vision
• Human Pose Estimation
• 3D Reconstruction

Background:

• Accurate 3D human pose estimation from images is crucial for various applications.
• Estimating poses of interacting humans presents significant challenges due to occlusion and lack of temporal information.
• Existing methods struggle with the complexities of close human-human interactions in sparse multi-view settings.

Purpose of the Study:

• To develop an automated method for detecting 3D poses of closely interacting humans from sparse multi-view images.
• To address challenges of partial occlusion, truncation, and lack of prior pose information in human-human interaction scenarios.
• To improve the accuracy and real-time performance of 3D human pose estimation for interactive scenarios.

Main Methods:

• Utilized OpenPose for 2D joint detection and Mask R-CNN for human semantic segmentation in individual images.
• Employed multi-view 2D joint triangulation to obtain initial 3D joint information.
• Developed a two-stage assembling method for accurate 3D pose selection and a novel approach to minimize interpenetration between human shapes.

Main Results:

• The proposed method successfully detects 3D poses of closely interacting humans from sparse multi-view images.
• Achieved a high visualized correct rate in experiments.
• Demonstrated superior accuracy and real-time capability compared to existing methods on the multi-view human-human interaction (MHHI) dataset.

Conclusions:

• The developed method offers an effective solution for 3D human pose estimation in challenging interactive scenarios.
• The approach significantly advances the state-of-the-art in accurately reconstructing poses of multiple interacting individuals.
• The method shows promise for real-time applications requiring precise understanding of human-human interactions.