Techniques for body and garment modeling

• Main Author: Zhang, Yu Zhe
• Other Authors: Nadia Magnenat Thalmann
• Format: Theses and Dissertations
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Computer graphics; DRNTU::Engineering::Computer science and engineering::Computer applications::Computer-aided engineering
• Online Access: http://hdl.handle.net/10356/63256
• Institution: Nanyang Technological University

The research documented in this thesis investigates techniques for human body and garment modeling. It aims to develop a collection of techniques that can be used to develop practical virtual try-on and made-to-measure applications in garment design and manufacturing industry. To this end, several new algorithms are developed, which include human body modeling, garment prepositioning, developable surface based garment customization, and topology editable free-form deformation (FFD). Specifically, our example-guided, anthropometry-based human body modeling method creates 3D human body models from users’ input of partial anthropometric measurements with a given example dataset. Rather than directly forming the mapping between the partial measurements and the body model, we first estimate a set of 30 chosen measurements from the partial input based on the example oriented measurement analysis. Then the example oriented radial basis function (RBF) model is established to map the set of 30 measurements to body shape space, and finally a constrained optimization is proposed to create the target 3D body model. The created model is guaranteed to match the input measurements and it reflects the shape characteristics of the examples as well. Garment prepositioning is a step to place 2D patterns of garment provided by manufacturers onto a 3D body mannequin properly for further modeling and simulation. It is crucial for automating garment design and virtual try-on. We present a solution to garment preposition, which consists of offline encoding and online decoding. The offline encoding places 2D patterns of the garment onto a template body and stitches the patterns into an initial 3D garment. The online decoding transfers the garment from the adapted template body model to the input body model. As a result, we can achieve automatic or semi-automatic garment prepositioning that works reliably on the input body in various poses. Moreover, we have also developed a Kinect based try-on application using the proposed body modeling and garment prepositioning techniques. The try-on application consists of four components: data extraction from Kinect, 3D body model generation, garment positioning and simulavition. This application demonstrates the usefulness of the proposed techniques. For a CAD designed garment which contains several 2D patterns, there is a need to modify the patterns such that the garment is customized for an individual customer. We present a garment customization method which takes 2D patterns as input and outputs adapted 2D patterns for manufacturing. The method goes through procedures from pre-fitting garment, developable surface based 3D garment customization, and 2D pattern flattening. In these procedures, both 2D and 3D criteria are considered: the measurements and shape of the 3D body, the developability of 3D garment, and the boundary and shape of 2D patterns. Spherical B-spline curve approximation is developed on the normal map. We optimize the developability and shape distortion of 3D surfaces in a global way. As a result, the proposed method can generate customized garment patterns that fit the individual body model and respect the original garment design. In addition to body and garment modeling, we also develop a general free-form deformation method in this research, which can be used to deform the shape of body and garment. The method is called T-FFD and is developed based on T-spline volumes. In this work, we introduce T-spline volumes into freeform deformation and design T-spline volume local refinement algorithm. The T-FFD offers several advantages over conventional FFD methods: (1) it allows local refinement of the control grid and supports editable topology of the control grid; (2) it provides a simple user interface in which the most effective control points are displayed; and (3) it offers a smart way to adjust the control points to perform the deformation.