1: \begin{abstract}

2:     In this note, we propose an approach to initialize the Iterative Closest Point (ICP) algorithm  to match unlabelled point clouds   related by rigid transformations. 

3:     The method is based on matching the ellipsoids defined by the points' covariance matrices and then testing the various principal half--axes matchings that differ by elements of a finite reflection group.   

4:     We derive bounds on  the robustness of our approach to noise and numerical experiments  confirm our theoretical findings. 

5: \end{abstract}

6: