3D skeleton construction by multi-view 2D images and 3D model segmentation. (2015)

• Record Type: Journal Article
• Authors: Tsai, Joseph C.; Chang, Shih-Ming; Yen, Shwu-Huey; Shih, Timothy K.; Li, Kuan-Ching
• Is Part Of: International journal of computational science and engineering. Volume 10:Number 4(2015)
• Page Start: 368
• Page End: 374
• Publication Date: 2015
• Subjects: automated clustering -- texture mapping -- SIFT -- 3D modelling -- 3D skeleton construction -- multi-view 2D images -- 3D model segmentation -- 3D objects -- shape diameter-function -- SDF -- Gaussian mixture model -- GMM; Computer science -- Mathematics -- Periodicals; Computer simulation -- Mathematical aspects -- Periodicals; Computational intelligence -- Periodicals; 004.015105
• Languages: English
• ISSNs: 1742-7185

Abstract

Issues regarding 3D object construction have been widely discussed for years. In order to simplify the process, we proposed a method to construct 3D object. Multi-view human images and feature points are used to generate 3D skeleton. We use an effective coordinate transformation method to transform feature points in 3D space. A modified K-means algorithm can search join points of target human by additional three directions and generate a simple 3D skeleton from the human's information. In 3D object segmentation, we use shape diameter-function (SDF) method and Gaussian mixture model (GMM). We use SDF method to compute the SDF value by centre of shape information and neighbour of current shape path information. The GMM method is used to obtain the scope value of object clustering in our paper. Eventually, we show results of our method in experiment results, and results show that our method is effective.