Answers are grounded in this paper and cite their sources.
Reliable infrastructure graph analysis requires transparent links between point-cloud observations, semantic annotations, spatial addresses, and graph decisions. This study evaluates G3H-GraphSim, a reproducible three-dimensional GeoHash cell-graph workflow on five adjacent Gwangju LAS–annotation tiles, using shared-origin metric grids, stage-level computational measurements, and tile-resolved paired routing statistics. Tile-local level-7 cells have unequal physical dimensions and largest-component shares of 26.0–95.1%. In a pooled grid with 4 m × 4 m horizontal cells, six-neighbor largest-component shares are 24.7% at 0.5 m and 26.5% at 1.0 m vertical resolution, showing that common addressing alone does not ensure graph continuity. A noncontemporaneous OpenStreetMap cross-check places 95.3% of retained road-cell centers within 10 m of mapped centerlines, although only 63.5% of selected reference-line length is covered. On the same 150 start–goal pairs, integrated routing reduces pooled mean speed-bump proximity and high-LiDAR-density cell entries by 18.0% and 28.5%, respectively, at a 0.6% increase in mean route length, with appreciable tile-to-tile variation. Stage-level tests show lower amortized batch exact-key lookup time and smaller key-array payload for packed integers than for padded Base32 strings. The results demonstrate traceable sensitivity to spatial resolution, representation, and routing policy, without establishing field-calibrated safety or deployed-system performance.
No takes yet. Share an insight, caveat, or question.
Kyusoo Chong (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: