Automatic preposition for an integrated virtual fitting solution using human model customisation technology

Abstract

The fashion industry has continuously flourished with the assistance of evolving technologies. Virtual fitting is one of these important technologies, which simulate the wearing effect of a clothing product on computers. For the customer end, virtual fitting technology facilitates clothing selection in online shopping as consumers can visualise the clothingfit before purchase. On the otherhand, virtual fitting technology also supports and streamlines the product development process of the fashion industry. Virtual fitting or clothing simulation is an active topic of research in the fields of computer graphics and fashion technology, although with different research focuses. The fashion technology community has the primary interest on simulation accuracy of virtual fitting. However, most existing virtual fitting solutions simulate clothing against a standard virtual mannequin, not on a 3D model with customised shapes and sizes of individual customers. This is partly because of the limitation of current human modelling technology, which tends to generate human models of average shape. Another reason is that most existing garment simulation solutions involve tedious user interactive operations to correctly position different clothing around human models before simulation. Moreover, the quality of simulation is only subjectively assessed, usually by visual inspections. There are no quantitative metrics defined to measure the simulation accuracy objectively. With an integration of the recent advancement in human model customisation technology, a new framework for intelligent virtual fitting is proposed in this research study. A prototype system is developed involving four integrated modules, namely human modelling module (HMM), pattern design module (PDM), garment simulation module (GSM) and garment rendering module (GRM). The key research developments of this study are mainly on GSM, in which preposition methods are proposed to automatically and accurately layout pattern pieces of different clothing designs in 3D space around human models of various sizes and shapes. Two garment simulators based on finite element model (FEM) a physical-based approach (FEM-GS) and a hybrid approach (Hybrid-GS) are constructed for automated garment simulations on customised human models. In addition, quantitative metrics are defined in this project to evaluate the simulation quality or accuracy. With the defined metrics, a comprehensive experimental evaluation is carried out to compare the two garment simulators FEM-GS and Hybrid-GS. By selecting suitable garment simulators, the accuracy of garment simulation can be ensured or maximised. Lastly, a workflow is suggested to the fashion industry for the selection of garment simulators by analysing clothing styles and fitting information.