A Method of Globally Optimal Registration for Multi-view Point Clouds Constrained by Closed-loop Conditions

Abstract

This paper proposes a rigorous registration method of multi-view point clouds constrained by closed-loop conditions for the problems of existing algorithms. In our approach, the point-to-tangent-plane iterative closest point algorithm is used firstly to calculate coordinate transformation parameters of all adjacent point clouds respectively. Then the single-site point cloud is regarded as registration unit and the transformation parameters are considered as random observations to construct conditional equations, and then the transformation parameters can be corrected by conditional adjustments to achieve global optimum. Two practical experiments of point clouds acquired by a terrestrial laser scanner are shown for demonstrating the feasibility and validity of our methods. Experimental results show that the registration accuracy and reliability of the point clouds with sampling interval of millimeter or centimeter level can be improved by increasing the scanning overlap.