Method for estimating real-scale 3D human body shape from an image based on 3D camera calibration and computer...

Area of Science:

• Computer Vision and Graphics
• Photogrammetry
• 3D Reconstruction

Background:

• Traditional 2D-to-3D reconstruction from low-resolution images is challenging without prior knowledge.
• Existing deep learning methods struggle with accurate real-world scaling of 3D human models from single images.
• Reverse projection photogrammetry is emerging as a replacement for combined computer vision and computer graphics analysis.

Purpose of the Study:

• To propose a novel method for estimating a real-scale 3D human body shape model (SMPL-X model) from 2D images.
• To overcome the limitations of existing methods in accurately scaling 3D models from single, low-resolution images.
• To enable detailed analysis of human posture, shape, height, and weight from standard camera inputs.

Main Methods:

• Utilized a combination of 3D camera calibration and computer graphics-based reverse projection photogrammetry.
• Developed a method to estimate the position, orientation, posture, and body shape of a 3D human body model.
• Integrated estimation of height and weight based on the reconstructed 3D human body shape.

Main Results:

• Successfully estimated real-scale 3D human body shape models (SMPL-X) from 2D images, even for non-straight postures.
• Achieved expanded analysis capabilities for height and weight estimation.
• Demonstrated the potential for person verification through evaluated equal error rates.

Conclusions:

• The proposed method effectively reconstructs real-scale 3D human body models from 2D images.
• This technique significantly enhances the accuracy and scope of human body analysis, including anthropometric measurements.
• The approach shows promise for applications such as person verification and detailed human modeling.