3D garment model reconstruction based on scattered point cloud

Abstract

We describe a method to quickly reconstruct 3D garment model. We first use a depth camera scanning device to quickly obtain 3D point cloud data on the surface of the clothing, pre-process the point cloud data. Then combining the region growth algorithm and Delaunay triangulation algorithm, we propose a surface reconstruction algorithm that takes undirected point clouds as input and generates interpolated surfaces in the form of triangulation to reconstruct the clothing model. We build this systematic 3D garment modeling program. Through our method we obtain a uniformly distributed grid, smooth patches, and also retain the original shape of the clothing point cloud. Our method increases the convenience of garment modelling, reduces the manpower and time for garment modeling, helps expand the garment database of the virtual fitting system, and provides ideas for garment electronic sales.

Status of research on parametric methods for the reconstruction of 3D models of the human body for virtual fitting

“…To more accurately reconstruct a model of the human body that matches its contours, Wang analyzed sequences of images in videos, applied the structure-from-motion (SfM) method to reconstruct Parameterized construction of 3D model a sparse lattice of the internal and external parameters of the camera corresponding to each frame of the image, relaxed the energy-related constraints and finally reconstructed a 3D model of the human body by using the SMPL-X model and multi-view fitting optimization. Figure 9 compares the results of reconstructions based on single and multiple views (Wang et al, 2021b). The left side of the above figure shows the results of reconstruction when multiple images were used and the right side shows those obtained when a single image was used.…”

“…Traditional parametric methods to reconstruct models of the human body use special equipment to obtain dense 3D point cloud data and then fit them with the parameters of SCAPE through point cloud registration, template deformation and other methods to reconstruct the 3D shape of the body (Wang et al ., 2021a, b; Han et al. , 2023).…”