[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]
Recently, 3-dimensional human body models are used in various area. In conventional methods, the model is considered as the set of body segments which are connected at joint parts. We call this model as ``joint-model''. These methods have insufficient abilities when we use them for the simulation of wearing clothes. The reason is that the methods can not represent the complicated shape deformation at the joint.
To solve this problem, a shape representation method based on the set of measured data of various poses has been proposed. In this method, the shapes of a joint part which correspond to the various poses are measured by a range finder. A lot of parameters are necessary in order to describe the variation of the measured data, which is caused by the change of pose, sufficiently. The problem is that the parameters are too many to control, and that the clustering method used here to reduce the number of parameters does not consider the distortion of the shape directly.
In this report, we propose a new method which can represent the complicated deformation of the shape at the shoulder joint and manipulate the pose easily. In this method, first, we acquire the shape data of various poses. Then, we manipulate the pose by using joint-model and fit a proper measured data to the manipulated joint-model. Here, we adopt the conventional joint-model which has a torso and an arm as two body segments.
In this article, we describe the procedure to estimate the model parameter values of the joint from a measured data in order to fit them to the manipulated joint-model. In this method, we acquire data of a shoulder with various poses using a range finder. The measured data are constructed as a mesh grid. When a group of vertices of the grid belongs to one body segment, we can assume that the motion of these vertices are represented by a single set of rigid body transformation parameters. Using this assumption, we can divide the vertices of the measured mesh grid data into the two clusters corresponding to the two body segments. As the basis of clustering, we introduce that of local rigidity.
After the clustering, we estimate the parameter values of the joint corresponding to the pose. As the parameter of the joint, we introduce a relative rigid body transformation parameter between the two clusters. Using the estimated parameter values, we can fit the measured data to the manipulated joint-model.
We experimented about the clustering and the estimation of the model parameter values of the joint . As the input data, we acquired the measured mesh data of various poses using the range finder, such as moving shoulder to forward and backward direction.
In clustering the measured data, it was shown that our method could cluster the mesh data more correctly than the conventional method. In estimating the parameter values of the joint, it was shown that the proposed method could apply to some types of pose change that the conventional method could not. From the experiment of fitting a joint-model with simple shape to the measured data by using the estimated parameter values, we confirmed the validness of the parameters.