A 3D avatar modeling of realworld objects using a depth camera

Abstract

In this paper, we propose a novel 3D avatar generation scheme using a depth camera. Depth camera can capture both visual and depth information of moving objects by using infrared light source at video frame rate. Our method consists of two main steps: alpha matting and mesh generation. We present a novel alpha matting algorithm that combines visual and range information. It improves an existing natural alpha matting methods. After alpha matting is performed a triangular mesh is created by using RGB, depth, and alpha images. Our method can represent any real world objects including a furry one. Experimental results show that ourmatting method demonstrates better results than the previous approaches. Especially our method provides a viable solution to model a scene with fuzzy objects.