Research Of Method And System For 3D Hand Modeling

In this paper a method to build a 3D model of human hand is proposed,which is extracting a 3D model as a template and superimposing pose transformation and shape transformation onto the template,by this way the model can be presented in a variety of different forms.The main contributions of this paper are as follows:First,registering different 3D hand models,which make the point-to-point distance between models can be calculated,and establish the standard for analyzing the difference between models.The process of model registration combines linear blinding skinning(LBS)and non-rigid ICP algorithms to eliminate the difference between models and make the models overlap together,and then use the KD tree algorithm to find the point-to-point pairing relationship between the two models.This method can be used in the registration of other types of 3D models too.Then,the parameter matrices required for 3D modeling of hand is generated,including the pose parameter matrix and the shape parameter matrix.The pose parameter matrix is used to compensate the skin collapse caused by LBS.The shape parameter matrix is splice by vectors extracted by principal component analysis(PCA)to ensure the template can be changed to different shapes.After these two matrices are generated,various 3D hand models can be generated by change the parameters of pose and shape vectors.Finally,a method of transforming 2D human hand image into 3D human hand model is proposed.This application is taken as an example to prove the usability of the model.Almost all of the problems above is solved by converted them into objective functions.The trust region method is used to solve these problems.