Monocular 3D Reconstruction of Human Body

Abstract

Human body 3D reconstruction using a monocular camera involves issues of geometry and appearance in the field of computer vision. A common idea of human 3D reconstruction is that fusion of point clouds collected by multiple RGBD cameras such as kinect v2 from different angles. However, depth cameras are more expensive than RGB cameras and they have stricter requirements of the environment. There are also many methods to reconstruct the point clouds of objects using multiple normal RGB cameras or monocular cameras. In this work, we propose a new application of Structure-from-Motion on human body 3D reconstruction. Structure-from-Motion is a process that the 3D structure is created by analyzing 2D image sequences, which is very prevalently used in many applications of computer vision. In our approach, we achieve human body 3D reconstruction for a 3D video conference using a monocular RGB camera.