How hkust-aerial-robotics/g3reg works
Project type and purpose
• G3Reg is a C++ open-source library and CLI tool for robust global registration of outdoor LiDAR point clouds. It provides a modular pipeline—from point-cloud loading to segmentation, feature extraction, matching, graph-based verification, and final pose estimation—configurable via YAML files and extensible via plugins.
File structure
• Top-level directories:
• /src: core implementation, split into front_end, back_end, utils, datasets
• Thirdparty: embedded external libraries (clique_solver, fpfh_matcher, robot_utils, backward-cpp)
• configs: parameter sets for different datasets and matching methods
• examples: demo and benchmark executables
• cmake: build tool modules
• docs: user guide, benchmarks, demo instructions
• Patterns:
• Modular separation: “front_end” for feature/descriptor, “back_end” for estimation/verification
• CMake‐driven build with find-package wrappers for Boost, Eigen, GTSAM, PCL, etc.
• Config-driven behavior (YAML files under configs/).
README for additional insights
• The pipeline graphic (assets/pipeline.png) highlights: segmentation-based matching with Gaussian Ellipsoid Model (GEM), distrust-and-verify via Pyramid Compatibility Graph (PAGOR), integration points for alternative matchers (FPFH, TEASER, RANSAC).
• Core features: fast GEM matching, robust PAGOR verification, generic plugin support.
Detail level
• Every module is annotated with its directory path (e.g., front_end/gem, back_end/pagor).
• Call out extensibility APIs (e.g., function signatures in utils/config.h, datasets_init.h).
• Configuration inputs (configs/*.yaml) feeding into each stage.