An approximate EM Homographical Iterative Closest Point algorithm

This paper describes an approximately expectationmaximization (EM) formulation of a homographical iterative closest point registration approach (henceforth HICP). We show that such an EM approach allows the algorithm to converge faster, and more robustly in the presence of noise. Although this algorithm can register points transformed by a more general set of linear transformations than the original Iterative Closest Point (ICP) algorithm, it is only appropriate for use on point sets which are related by a homographical transformation, e.g. images taken of a planar scene from different angles, or images taken of a general scene by a stationary pan-tilt-zoom camera. The algorithm is tested on real and synthetic data.