Cross-Parameterization and Its Applications in Customized Design.
详细信息
文摘
Human-centric product design is a challenging work； this is because the products and human bodies are in freeform shapes. The design customization for a particular human is generally taken in an interactive manner. However,this is very time and labor consuming. If a previous design can be reused for a new human model,a lot of time and labor works are saved. To enable the design reusing,a bijective mapping should be found for different human models. Meanwhile,such mapping should have low distortion,so that the reusing result is practical. In this thesis,surface and volumetric cross-parameterization are presented to find the bijective mapping among the human models. The cross-parameterization is based on the domain decomposition method,which can interpolate the semantic features of human bodies. A robust domain construction algorithm is developed based on the Voronoi Diagram respecting the semantic features. The domain constructed by this algorithm is the best triangulation under the hard constraints,and the mapping computed by this domain is having low distortion. Furthermore,an efficient domain optimization algorithm is presented to further optimize the domain for different models at the same time. A new signature,Length-Preserved Base-Domain LPBD),is investigated for measuring the level of stretch between domains in cross-parameterization. Based on LPBD,the domain connectivity is optimized efficiently. Experiments show that the optimization can effectively reduce the parameterization distortion. The surface cross-parameterization is further extended to volumetric cross-parameterization in order to define the mapping around the human bodies. The surface domain construction algorithm needs to be elevated to one more dimension. To enable this elevation,an intersection-free patch construction algorithm is developed to decompose a volume into a set of sub-domains. The parameterization distortion is minimized by two stretching operators. Before computing the mapping,one important problem for designing human-centric products is the human body reconstruction. Human models are needed for doing any designs or reusing. Recently,RGB-D cameras are used for human body reconstruction； this is because not only they can provide depth information,but also these cameras are low-cost and fast. Comparing to traditional human body scanner -- laser scanner,the one using RGB-D cameras is more ready to be used out-of-laboratory. In this thesis,an "instant" human body scanner is developed for data collection,and a volumetric template fitting method is presented to reconstruct human bodies from incomplete data. The techniques proposed above are good for the products with low-customization cost e.g.,clothes,furniture,etc.). For those products that are not made by planar materials,customization for all components is too expensive. On the other hand,the electronic and mechanical components cannot be freely varied. In order to solve these problems,human-centric product design using standardized components is requested. The standardized component is a component that associated with an element library,and the element library is a serial of the components variations that can be manufactured by mass production. In this thesis,a Mixed-Integer As-Rigid-As-Possible MI-ARAP) shape optimization framework is developed for automating the customization of human-centric product. This framework can automatically select the best-fit standardized components from the element library,and the shapes of the standardized components are preserved in the optimization. Experimental results prove the success of this framework.