Autonomous navigation for vehicles in non-trivial environments requires fast and precise localisation. Autonomous vehicles typically operate in 3D environments and require 6DOF localisation, which is computationally costly. We have developed fast, accurate localisation by constraining the 6DOF search space by physical vehicle limitations, and by the use of Graphics Processing Units (GPU's) to enable the use of dense 3D internal representations of the environment. This paper presents a method of selecting sub sets of polygons from the 3D map for use in localisation, in order to further reduce computational load, while maintaining localisation accuracy. Polygons are indexed by spatial location and a subset is selected by ranking visible polygons by observation frequency, angle of incidence to sensor ray and likelihood to fall within the sensor field of view. Pre-computed sets of polygons are then used to perform 6DOF localisation. An autonomous navigation experiment at a power sub-station facility is presented. Localisation accuracy is maintained while computation costs are reduced by a third.
展开▼
