Chapters
21 of 21 entries shown
Foundations and Acquisition
This part defines point clouds, explains how LiDAR measures range and where its errors arise, and examines the acquisition platforms, georeferencing, file formats and spatial indexes that affect every later operation.
Chapters 1 to 4
- 01Introduction to 3D Point Clouds FreeWhat a point cloud is, the six sources it comes from, the error budget every product inherits, and the pipeline the book follows.
- 02LiDAR Principles and Technology FreePulsed, phase-shift and FMCW ranging, the link budget, scan mechanisms, intensity, and the georeferencing chain with its error budget.
- 03LiDAR Acquisition Systems FreeSpecifying an acquisition, reference frames and survey points, fixed, moving and image-based platforms, mission planning, and georeferencing checks.
- 04Data Formats and Spatial IndexingLAS, LAZ, E57, PLY, COPC and EPT, the point record, k-d trees, octrees and grids, compression, tiling and out-of-core processing.
Preprocessing
Preprocessing prepares observations for analysis: sampling reduces the point count, filtering identifies unwanted returns and reduces noise, normal estimation describes local surface direction, and registration aligns datasets in a common frame.
Chapters 5 to 8
- 05Sampling and ResamplingRandom, voxel grid, farthest point and Poisson disc sampling, normal-space and adaptive variants, and what each one keeps and loses.
- 06Filtering and DenoisingRange noise, outliers and mixed pixels, outlier removal, surface projection, edge-preserving smoothing, learned denoisers, and how to check a cleaning pass.
- 07Normal Estimation and OrientationPCA plane fitting, neighbourhood size, curvature from jet fitting, robust estimates at edges, learned estimators, orientation, and diagnostics.
- 08RegistrationClosed-form alignment, ICP with its point-to-plane, GICP and NDT variants, coarse registration by descriptors, pose graphs and SLAM, and the evidence for accuracy.
Analysis
Analysis converts local geometry and attributes into useful descriptions by separating ground from other returns, estimating geometric features, grouping points into regions, assigning semantic classes and establishing correspondences between scans.
Chapters 9 to 13
- 09Ground Extraction and DTM GenerationDTM, DSM and nDSM, the two errors every ground filter makes, slope, morphology, TIN and cloth simulation filters, tuning by site, and hydrological consistency.
- 10Local Geometric FeaturesThe seven eigenvalue features, roughness, verticality, height and density, density bias and edge contamination, multi-scale vectors and scale selection.
- 11SegmentationRegion growing, RANSAC, supervoxels, graph cuts and density clustering, thresholds read off the data, and how to measure a partition.
- 12ClassificationConfusion matrices, IoU and kappa, the feature vector a classifier receives, spatial splits and class imbalance, tree ensembles to learned representations.
- 13Local Descriptors and Feature MatchingKeypoint detection and repeatability, histogram, spin image, frame-based and learned descriptors, matching and verification, and place recognition.
Reconstruction and Semantic Modelling
Reconstruction and change analysis derive geometric products from sampled observations by comparing survey epochs, fitting surfaces, reconstructing objects such as buildings and trees, and organising their geometry and meaning in semantic models.
Chapters 14 to 17
- 14Change DetectionCloud-to-cloud and cloud-to-mesh distance, M3C2 with its level of detection, volumes, displacement and strain, and time series from permanent scanners.
- 15Surface Reconstruction and MeshingHeight fields, alpha shapes, ball pivoting, greedy projection, Poisson and screened Poisson, radial basis functions, signed distance fields and Marching Cubes.
- 16Object ReconstructionRoofs, façades and levels of detail, trees as cylinder hierarchies, roads, conductors, rooms and pipe runs, and specifying a testable object model.
- 17Semantic 3D Models and Digital TwinsGeometry, semantics, topology, georeferencing and provenance, valid solids under ISO 19107, levels of detail, CityGML, CityJSON and IFC, and national building models.
Applications and Frontiers
Applications such as national mapping, forestry, heritage, mining and autonomous driving determine which products are needed and how their quality should be assessed, and the final chapters examine deep networks, radiance fields and Gaussian splatting.
Chapters 18 to 20
- 18Applications of 3D Point Cloud Processing FreeNine domains on one map of extent, accuracy class and repeat interval, from urban models and driving perception to forestry, heritage, bathymetry and mining.
- 19Deep Learning for Point CloudsPointNet, hierarchical grouping, point convolution, graphs, attention and fusion, detection and instance segmentation, training signals, and the domain gap.
- 20Radiance Fields and Gaussian SplattingNeRF and Gaussian Splatting, surface-constrained variants, LiDAR inside the optimisation, and compression, viewing scale, dynamics and city-scale extent.
Appendix
The appendix develops the mathematical tools used throughout the book: vectors and distances, rigid transformations, principal component analysis and least-squares estimation with uncertainty propagation.
All 20 chapters and the appendix, online and as the 840-page PDF, with free updates.