A self-calibrating registration system for arbitrary-view 3D scanning

Abstract

Three-dimensional (3D) measurement technology has been widely applied across various industries, yet the challenge of registering multi-view point clouds into a unified coordinate system persists. In this paper, we propose and mathematically derive a flexible calibration model for an arbitrary-view registration system, which consists of a real Binocular Structured Light Measurement (BSIE) unit and a turntable. When the turntable rotates, one real and multi-virtual measurement units are distributed along a spatial circle. The proposed method first constructs a model describing the positional relationship between the real and virtual units relative to the turntable, with the model parameters calibrated using a single 2D calibration target. Secondly, the data of point clouds of the object surface in different views are measured by BSIE units. After that, the set of point clouds is aligned into the global coordinate system utilizing the calibrated model parameters. So far, the point cloud registration is completed for regular objects with limited texture information. For complex surfaces, ICP is applied as a post-processing step to refine results enabled by accurate calibration. Furthermore, when the positional relationship of the system changes, the calibration parameters will no longer be applicable. In order to improve the convenience of the algorithm, some marker points are set on the turntable to realize the automatic calibration. Experimental results on two objects with regular and complex shapes demonstrate that the proposed method is convenient, fast, and easy to calibrate, with a registration error of less than 0.08 mm.