Linear quaternion interpolation reveals globally consistent closed-circuit point cloud registration

Abstract

In the task of 3D environment mapping from point cloud registration, we face the problem of global consistency analysis in the presence of noises and a low overlap rate. Significant errors in the registration result in less consistent environment maps. This study aims to present a linear interpolation method for consistent terrestrial laser scanning data registration to overcome the drift problem from pairwise registration errors, taking advantage of loop closing detection into a circuit. A coarse-to-fine point cloud registration is constructed beforehand. Then, a linear interpolation approach is proposed to redistribute the registration errors further without constraints globally. Compared with the mainstream point cloud registration algorithm, the proposed method shows the highest global consistent accuracy in different scenarios. Our method achieves state-of-the-art with global registration accuracy within 5 mm, sufficient to create compatible 3D maps.

Bibliographic Information