Methods for generating scan poses to optimally complete partially acquired 3D scenes [Methoden zur Bestimmung optimal ergänzender Scanposen in unvollständig erfassten 3D-Szenen]

Abstract

This paper describes and evaluates several methods to create an optimal set of scan poses to complete a given partial model of a scene. Random generation of pose candidates is compared to an approach that actively creates good candidates with respect to clusters that are extracted from the environment. To reduce the computational cost of scan simulations by raytracing, an inverted model called reverse raytracing is developed and used for evaluating the poses and choosing an optimal subset, either by a greedy strategy or by describing the problem as an Integer Linear Program that not only offers an exact solution but also a faster approximation. Further hardware restrictions such as vertical/horizontal angular aperture and minimal/maximal height of pose candidates are considered in order to adapt the program to the particular situation.