Robust generalized total least squares iterative closest point registration

This paper investigates the use of a total least squares approach in a generalization of the iterative closest point (ICP) algorithm for shape registration. A new Generalized Total Least Squares (GTLS) formulation of the minimization process is presented opposed to the traditional Least Squares (LS) technique. Accounting for uncertainty both in the target and in the source models will lead to a more robust estimation of the transformation. Robustness against outliers is guaranteed by an iterative scheme to update the noise covariances. Experimental results show that this generalization is superior to the least squares counterpart.