Methodology for the Evaluation of Point Cloud Maps for Absolute Positioning - Quantitative Investigation of the Accuracy of Environmental Representation

Kurzfassung

Reliable local accuracy information is difficult to achieve for large point-cloud maps because high-quality reference observations are usually sparse. This thesis presents an automated offline pipeline for deriving a spatially resolved horizontal correction field from surveyed pole landmarks. The pipeline preprocesses the point cloud, detects pole footpoints, georeferences them in a common terrestrial reference frame, estimates a robust global two-dimensional alignment, and establishes one-to-one correspondences with a reference catalogue. The remaining correction vectors are then interpolated using uncertainty-weighted inverse distance weighting (IDW) and evaluated through leave-one-out cross-validation. The evaluation covers 14 configurations from four urban routes, combining RTK and non-RTK recordings with LIO-RF and FlexCloud georeferencing. In 12 of the 14 configurations, IDW achieved a lower held-out RMSE than global alignment alone and outperformed nearest-neighbour correction in all 14 configurations. Its benefit depended strongly on the spatial structure of the remaining deviations and the availability of nearby reference matches. The largest relative improvement over global alignment alone was 52.3%, whereas no improvement was observed for a configuration without significant positive spatial autocorrelation. These results show that local interpolation can improve positional agreement where systematic spatial variation and sufficient reference support are present. The resulting grid should therefore be interpreted as a conditional correction field with diagnostic uncertainty and support information, rather than as an unrestricted guarantee of map accuracy.