Heuristicallly targeted Minimum Description Length test for stone detection from public point cloud data

Coarse cross-terrain point clouds are gathered by aerial laser scan (ALS) and dense point clouds by unmanned vehicle (UAV) operation. These two data sources have complementary nature and should be combined for various applications. This paper uses minimum description (MDL) length approach to detect individual stones and their physical dimensions from UAV data. The MDL procedure is spatially targeted by a two-step heuristics: local stoniness likelihood derived from ALS data and the curvature detection on UAV data. Comparison of the performance of MDL principle and a geometric approach, namely mean square error (MSE) minimization is presented. The MDL approach can be applied to cloud point densities ρ ≥ 3 m−2.