| （英） |
The fusion of computer vision (CV) and computer graphics (CG) is progressing in many fields. For example, in the forensic science domain, reverse projection photogrammetry, that identify geometric properties (position and shape, etc.) of a subject based on accurate reproduction of the image content, are often used to measure height, etc.; Application to a human image is a challenging task and has not yet been realized. On the other hand, in recent years, many researches have been focused on deep learning-based technology that applies fitting a 3D human body shape model (SMPL model, etc.) to an image, but it is difficult in principle to scale the reconstructed model to actual scale. In this study, we propose a method to estimate a real-scale 3D human body shape model (SMPL-X model) from a human image based on the combination of 3D camera calibration and CG-based reverse projection photogrammetry. In addition to evaluating the degree of difference between original silhouette image and estimated silhouette image by using the proposed method, it becomes possible to estimate position, orientation and posture of a 3D human body shape model of a human image of not standing upright, which is conventionally difficult to be anlyzed. Based on the estimated 3D human body shape, estimation of height and weight becomes possible, the range of analysis of height/weight estimation largely expanded. In the near future, we aim to expand the proposed analysis to one-by-one person verification task. |