3D Reconstruction of the Human Body from Partial Scans Using Parametric Models

This paper aims to reconstruct complete 3D models of the human body from partial scans obtained using low-cost RGB-D cameras. These cameras provide partial views in the form of noisy point clouds of body parts. To address the complete 3D reconstruction of the body from these partial views, parametric models such as SMPL and STAR are used, whose parameters determine the pose and shape of the bodies obtained. A complete pipeline is proposed that addresses from the preprocessing and segmentation of partial clouds at the entrance, going through the parametric optimization of pose and shape, as well as the integration of the different models obtained. The results of the reconstruction of the body models are shown, which in qualitative terms faithfully represent the expected shapes. At the quantitative level, the distance metrics of the points obtained in the models with respect to the “Ground Truth”, show promising results, in relation to other methods that start from complete scans.