[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]
In CG domain, effort has been made to calculate and simulate shapes and deformation of objects in the real world on computers. Target objects were limited to rigid body to recent years. Since geometrical and physical characteristics of flexible objects are far complicated from those of rigid objects, simulated 3-dimensional behavior of flexible objects must reflect those complicated characteristics. Therefore it takes substantial time to calculate behavior of rigid objects. Today we can deal with flexible objects because of hardware evolution and development of techniques.
In order to describe shape of cloth and to calculate a deformation against time, it is necessary to build a cloth model on computer. Cloth modeling methods include geometrical approach and physical approach, and there are also hybrid methods recent years.
On decision methods of model parameters expressing characteristics of cloth, on the other hand, a method where a user adjusts parameters based on human experience has been generally used to get shapes which agree with shapes in the real world.
Due to complicated characteristics of cloth, it takes considerable time to acquire parameters with this method that can simulate various appearance well. Thus a technique is invented that measures physical characteristics of cloth and directly reflects them to a model, and is used widely today. Since this technique takes actual cloth characteristics in, model parameters acquired are expected to be able to correctly simulate wide variety of cloth shapes and their deformation. This technique, however, requires special measuring devices and it takes huge time to calculate parameters. Moreover, in case of focusing objective on cloth shape simulation, the method to measure dynamics characteristics is not wise policy because not all physical characteristics may be necessary.
This study aims to obtain parameters of cloth model which reflect actual cloth characteristics to some extent and can simulate various shapes without measuring physical characteristics of cloth. To this end, we observe several shapes in a comparative simple device, and by tuning model parameters so that this model has similar shapes to those observed shapes, we simulate various shapes including observed shapes.
We use a Motion Processor, a simple device which can observe shapes and their deformation of objects against time. In this paper we propose a method to obtain model parameters which afford to reproduce various still shapes using still shape data observed by the Motion Processor.
We adopt mass-spring model, a general physical model, and define two energy functions, representing difference between observed shapes and model shapes and force inballance, to make the model simulate observed shapes as still shapes, then apply steepest descending method to these functions to get parameters such that sum of functions has local minima. Lastly we compose the cloth model.
In order to validate our method, first we experimented reproduction of observed shapes by obtained model. Second we experimented reproduction of foreign shapes by obtained model. As a result of our experiments, we verified that the model obtained by the method can simulate observed shapes well and foreign shapes which has fair accuracy.
As a result of our experiments, we have proved that, using a motion processor, we can simulate various complicated shapes of real cloth without measuring actual physical characteristics of cloth.