Estimating Rigid Transformation Between Two Range Maps Using Expectation Maximization Algorithm

Rigid registration of two sets of points sampled from a surface has been widely investigated (e.g., [1], [4]–[6], [9]) in computer vision literature. Generally, these methods are designed to tackle range maps with dense points for non-realtime applications. In [2], [8] scans are matched using iterative closest line (ICL), a variant of “normal-distance” form of ICP algorithm [1] originally proposed in computer vision community by [3]. However, the convergence of this approach is sensitive to errors in normal direction estimations [10].