The generic surface reconstruction of the previous chapter is a single method that produces a faithful triangle mesh of whatever the points happen to describe, and it treats a building, a tree, and a powerline span as three instances of one problem. Reality does not. Each of those objects obeys its own structural grammar, and the method that ignores those grammars throws away exactly the structure the reader will later need. A roof is a small set of planar facets meeting at ridges and eaves. A mature tree is a hierarchy of tapering cylinders that TreeQSM (Raumonen et al., 2013) fits to stem-volume errors of 5-15% on calibrated specimens. A powerline is a catenary parametrised by tension and span, fitted on cleared ALS corridors to sub-decimetre residuals. The cost of staying generic is that no consumer, from the city cadastre to the forestry inventory to the utility maintenance scheduler, can ask the right questions of the model.
Object reconstruction draws on more of the pipeline than any chapter before it. The classified clouds of Chapter 12 supply the per-class points, the instance segmentation of Chapter 11 isolates the individual objects, the ground surfaces of Chapter 9 anchor heights and footprints, and the eigenvalue features of Chapter 10 flag the planar and linear structures that the fitting stages exploit. The structured models it produces feed the city-scale, forestry, and infrastructure workflows of Chapter 17, and Chapter 18 revisits several of its subproblems with learned predictors. Whatever the domain, the work follows the four-stage skeleton shown in the figure below: classify the points, isolate the object, fit the model, and export the structured result. The chapter proceeds domain by domain, through buildings, trees, roads, powerlines, façades, indoor environments, and industrial infrastructure, and closes with a comparison of methods across accuracy, robustness, and automation level.
Building reconstruction from LiDAR data is driven by the demand for 3D city models at various levels of detail, standardised by CityGML as LoD2 and LoD3.
The CityGML standard defines a sequence of levels of detail (LoD) for 3D building models, where each level adds geometric and semantic richness over the previous one:
Most ALS-based workflows target LoD2, while TLS and MLS data enable LoD3. The figure below shows the five levels side by side on the same building.
You have just read the opening section. Pick one and keep reading.
Read all 19 chapters online here, and download the complete PDF on release.
Everything in the book, plus the companion video course as it ships.
By Abderrazzaq Kharroubi, geomatics engineer: a decade of LiDAR fieldwork, 2,000+ students taught. One payment, lifetime access, 30-day refund.
Anything you highlight or note stays with you after you buy. Compare all tiers or sign in if you already bought.