Generating 3D Virtual Garment Based On 2D Sketches

The technique involved in creating three-dimensional (3D) virtual garment is to simulate the performance of real fabric and garments in real world, which covers the state of the art from both textile engineering community and computer graphics society.The research on 3D virtual garment can be categorized into several groups:generating 3D virtual garments based on virtual sewing method, using parametric surface patches to shape 3D virtual garments, constructing 3D virtual garment based on constrained contour curves and stylish curves, and sketch-based 3D virtual garments construction. Among which, the first 3D garments generating method is relied on physical modeling of clothing. The major steps including sewing, collision detection/response, and strain control to maintain the size stability. While the last three construction methods are based on geometric modeling, i.e. the 3D virtual garments are obtained by changing the geometric structures of surface meshes.Although these methods can generate plausible visual effects and animations, it is not feasible for a designer to employ such pipeline in his/her daily design routine, since the natural way of garment design is usually conducted via pencils and papers, i.e., a sketch based approach, In this paper, we present a system to fulfill the requirements of a garment designer. The system is designed by interpreting 2D garment sketches to 3D virtual garments isomophically.Firstly, an interactive interface is constructed to support 2D garment sketching. Our job in garment sketching identification is mainly relied on uniform quadratic B-splines, i.e. B-splines are used to fit 2D strokes. In this system, the original design is defined in front and back panels. The 2D garments are then mapped onto the 3D garment-templates in terms of the Euclidean distances between 2D points and 3D vertices. In this sense, the 3D garment-template models are employed as a bridge between 2D gannent sketches and 3D final garments in shape interpreting. Given a distance threshold, the matching vertices on 3D garment-templates surface mesh can be found accordingly. The interactive sketching interface is feasible for users as it provides various editing tools including adding/deleting control points, deleting single stroke, modifying/moving control points and saving the entire sketches.In our practice, the 3D template models are obtained from various sources including 3D garment CAD system,3D shape authorization software and range scanned data. The 3D template models are of various mesh quality, either regular or irregular. Mesh regularization is then inevitable to reach a satisfied result in the future surface editing to add wrinkles and folds.The procedure of 3D mesh regularization can be divided into two steps. The first step is to resample the original surface through a slice-by-slice ray/triangle intersection. The second step is to reconstruct the surface with the resampled vertices through a ball-pivoting algorithm, which is derived from 3D alpha-shapes reconstruction. Since the ball-pivoting algorithm is to shape an interpolated surface in a point-based manner,, the reconstructed surface and the original template surface are in the same manifold.After mesh regularization, a matching scheme between 2D sketches and 3D template models is performed. Technically, correct matching guarantees correct mesh deformation. The matching is implemented on both front and back panels with an initial treatment followed by a direction-driven matching scheme.The initial treatment is to find the corresponding vertices on 3D template mesh for each 2D strokes. The correspondence is set up based on distances between 2D points on strokes and 2D vertices on 2D garment panels. The 2D garment panels are obtained by setting z= 0 for front and back panel vertices of 3D garment-templates. Given a distance threshold, the initial matching vertices can be found by this way. A linked list is employed to speed up the sorting/deleting in this process, and to remove the possible redundancy such as many-to-one during the matching.However, the initial matching vertices can not form a continuous curve on 3D surface. Hence an advanced matching method based on vector orientation is employed to find other vertices on the curve. In our practice, we found this direction-driven matching is suitable for surface meshing with uniform topology, and is fast for acquiring correct matching between 2D strokes and 3D garment-templates.The last part of our research is focused on wrinkles and folds generation, which belongs to the research area of mesh editing.. Loop's subdivision, Laplacian mesh optimization, Mean-value encoding/decoding, and surface smoothing are discussed in this part.Since a triangle patch after Loop's subdivision can be divided into 4 sub-patches, and the subdivided mesh is in C2 smoothing, the surface mesh after Loop's subdivision is uniform, smoothing and compact. The boundary of reconstructed surface of 3D garment-templates is in C1 smoothing. The detailed wrinkles can be shaped after Loop's subdivision.Mathematically, both Laplacian deformation and mean-value decoding are locally defined deformation, i.e., the new position of the vertices is computed based on local geometric information. The local property is to retain the angels and lengths between a vertex and its neighboring vertices. When the matching vertices are moved to target positions, the mesh topology is retained as much as before. Therefore these two mesh deformation techniques are suitable for garment deformation based on templates..