• Volume: Once tiled, all the processing steps involved only consider the whole data at the tile level and constant size tile-metadata at the global level: data processing services are then only processing a constant number of tiles at the same time, possibly in a streaming fashion.
• Variety: The proposed approach is able to merge datasets from various sources, taking into account their different levels of uncertainty and geometric setups (terrestrial or aerial lidar, photogrammetric point clouds...)
• Velocity: To ensure rapid processing, exactness has been only ensured when it impacted reasonably on the resulting performance (or tractability). As such, the Delaunay triangulation has been kept globally exact over the whole data set, as well as in the embarrassingly parallel pre-processing steps (optional resampling, local geometry dimensionality analysis) and in the merging of input datasets into scoring values supported by the Delaunay triangulation, whereas the extraction of the final surface from the score-valued 3D Delaunay triangulation is not guaranteed to provide the globally optimal solution, which is intractable. Instead, a parallel tiling & stitching solution is employed to efficiently compute close-to-optimal results.
• Veracity: The watertightness property could have come at the expense of hallucinated parts in the result, just to enforce this property. The per-triangle confidence measure provided intrinsically by the proposed method clearly states the level of veracity of each part of the resulting triangular mesh, such that the user (either a human or a machine) can assess immediately the appropriate level of trust in the model accuracy.

The figure shows the unified reconstruction by merging aerial and mobile mapping LiDAR point clouds. The coloring denotes the estimated confidence: red-high uncertainty from aeral samples, grey- low uncertainty from mobile mapping samples.