Two-step registration in 3D human body scanning based on multiple RGB-D sensors
In this paper, we propose a universal registration method for the body scanner consists of multiple fixed sensors. Firstly, a new double-faced checkerboard is designed to apply the calibration method into the initial estimations of the relative poses of point clouds. Secondly, ICP (Iterative Closest Point) algorithm is utilized for further registration. Since the ICP algorithm may fail in situations of outliers, ambiguity, or missing data, we make full use of the geometry feature of creased papers to solve these problems. A new body scanning system with fast capture speed is also proposed to validate our method. The experiment results demonstrate our approach is robust and fast.
PDF