|Place of Conferral||北京|
|Keyword||虚拟试穿 骨骼动画 穿透修复 刚性配准|
The Mass-production of clothing is still the current status of fashion industry, which prefers to produce garments based on pre-defined pattern sizes(e.g. S, M, L, XL and XXL). Customers choose the great-fitting one by trying on different sizes of clothes while virtual try-on(VTO) systems try to finish the fitting process under the virtual space via 3D body and garment models. This thesis concentrates on size-preserving garment transferring and fitting and a physically-based framework, consisting of two stages: rough garment-body matching as the global registration and garment-body penetration resolution as the refinement, is proposed. Furthermore, the two stages can be subdivided into several steps, including rigging static human meshes, posture adjustment and alignment, rigid registration between body and garment models as well as quantitative assessment for fitness.
1) Automatic character rigging and skin attachment. Rigging a static mesh is far from a solved problem. Thus, we adapted the state-of-the-art techniques to meet these requirements of VTO. To be specific, for human bodies in symmetric poses, Pinocchio system gives a partial solution. Taking advantage of its skeleton embedding and rigging techniques, we make it applicable in our garment fitting framework. For human bodies in unsymmetric poses, we make some improvements upon existing automatic joints extraction algorithms based on Reeb Graph, including more reasonable heuristic rules as well as exploiting cluster methods to guarantee more credible joint positions.
2) We propose a two-stage algorithm, swing alignment plus twist compensation, to perform skeleton posture alignment. After that, the pose of the target body is aligned with that of the reference body with skeleton-driven deformation, which further allows us to transfer garment models onto the target body via rigid transformation. To be specific, we treat two human models as two sets of point clouds and exploit the rigid ICP registration to derive the appropriate affine transformation, which is directly applied to the garment models.
3) We further improve the penetration resolution algorithm based on the minimum positional displacement(MPD) and make it more appropriate for resolving pre-existing cloth/body penetrations in the scene of VTO. It is history-free and uses discrete intersection detection(DID) to find out all intersecting geometric primitives, and relies on the body surface orientation to identify vertices that are in illegal state. These illegal vertices are relocated, with minimum displacements, so that the vertex-to-face signed distances are zeroed. MPD outperforms existing methods for handling penetrations between bodies and clothes, and is particularly effective in contexts like virtual-try-on and example-based garment animation synthesis.
Compared with the existing VTO frameworks，our proposed scheme has the following superiorities: (1) The fitting process is completely automatic for human bodies in symmetric poses, with no need for extra manual tweaking; (2) Garment transferring between bodies with arbitrary shapes as well as arbitrary poses is well supported. (3) Only elastic deformation is allowed with dimension preserved during the fitting pipeline. (4) The results of garment fitting satisfy the laws of physics at the most extent.
|孙立明,孙立明. 尺寸不变的服装迁移及虚拟试穿方法研究[D]. 北京. 中国科学院大学,2016.|
|Files in This Item:|
|201328014629078孙立明fi（24916KB）||学位论文||暂不开放||CC BY-NC-SA||Application Full Text|
|Recommend this item|
|Export to Endnote|
|Similar articles in Google Scholar|
|Similar articles in Baidu academic|
|Similar articles in Bing Scholar|
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.