"Corrected Perth CBD Point Cloud Data 2021"

"This dataset is a post-hoc geo-registered (calibrated) release of the Perth Central Business District (CBD) high-definition LiDAR map originally captured by Ibrahim et al. (DICTA 2021) [1]. The raw map was constructed from closed-loop terrestrial LiDAR scans without any GNSS or IMU input, so it lives in an arbitrary local Cartesian frame with no global georeference and cannot, in its original form, be overlaid on commercial maps or imported into a GIS.This release supplies the missing geographic frame. Every point has been re-projected into UTM Zone 50 South (EPSG:32750, MGA94) using the four-stage calibration pipeline described in our companion paper [2]: semantic road segmentation with Point Transformer V2, satellite-imagery-driven keypoint matching, global rigid alignment by similarity transform, local non-rigid refinement with thin-plate-spline RBF interpolation, and absolute vertical correction against SRTM 30 m terrain.The corrected cloud contains 27,656,188 RGB-coloured points covering \u2248 3.0 km \u00d7 1.4 km of the CBD. Calibration accuracy has been independently validated against four map providers (Google, OpenStreetMap, Mapbox, HERE) and SRTM. After correction, the merged planimetric error against Google road centrelines drops from 13.35 m \u2192 3.01 m; vertical error against SRTM drops from 13.55 m \u2192 3.91 m, with elevation correlation rising from 0.38 \u2192 0.86. Per-region planimetric residuals are as low as 0.52 m (CBD-03), and cross-source residuals (2.5\u20134 m) sit within the inter-provider uncertainty itself, confirming the calibration is not over-fitted to any single reference.The contribution of this dataset is the calibration. The underlying LiDAR geometry is unchanged from Ibrahim 2021 [1]; what is new is the rigid + non-rigid + elevation transformation that turns it into a georeferenced product directly usable in GIS (QGIS, ArcGIS), web visualisation (Cesium, Potree), and CAD (Rhino, MeshLab, CloudCompare). The dataset is intended for benchmarking GNSS-denied geo-registration of terrestrial LiDAR, urban 3D mapping and digital-twin construction, cross-modal alignment between LiDAR and satellite imagery, and vehicle-localisation experiments in dense urban canyons."