Spending way too much time manually classifying point clouds in TerraSolid. Ground, vegetation, buildings, noise... it takes hours on large datasets.
Are any of the AI-based classification tools actually good? Heard about tools from Terrasolid's own AI module, Bentley's iTwin, and various academic solutions.
Anyone using automated classification in production and trusting the results?
Sign in to join the discussion
Sign inTerraSolid's machine learning classifier has gotten really good. Train it on a manually classified tile and it does 80-90% of the work on similar terrain. Still need manual cleanup but it cuts processing time by 60-70%.
The key is training on representative data. Forest terrain trains differently than urban.
CloudCompare with the CSF (cloth simulation filter) plugin does a solid job of ground classification for free. Not as polished as TerraSolid but for basic ground/non-ground separation it's 90% there.
For building/vegetation separation, PDAL's SMRF and PMF filters work reasonably well too.