Boosting the Computational Performance of Feature-Based Multiple 3D Scan Alignment by iat-k-means Clustering

In this work we present a method to control and cut down the computational time required by feature-based multiple-view alignment solutions employed in modern 3D modeling pipelines. The reduction of the number of feature matches is guaranteed for each added view by means of an incremental (allowing dynamic views addition) and adaptive (variable number of clusters) implementation of a k-means clustering. The proposed method also comprises convergence quality and cluster cardinality control mechanisms, and guarantees multiple view alignment in nearly constant time with respect to the number of scans that need to be aligned for a significant class of feature descriptors. Moreover we demonstrate, on a representative experimental dataset, that the per-view alignment time can be reduced to a fraction of the corresponding pair wise alignment time without any performance degradation in terms of successful alignment. The obtained results are relevant for several 3D modeling applications where, especially for the acquisition of big and complex datasets, automation and robustness requirements are to be coupled with a quick and interactive usage of modern range scanners.