Commit ae9521f
feat: implement fine registration algorithms (ICP variants and NDT)
Implemented 5 fine registration algorithms for precise point cloud alignment:
- Point-to-Point ICP: Classic ICP using SVD for transformation estimation
- Point-to-Plane ICP: ICP variant using surface normals for better convergence
- Generalized ICP: Plane-to-plane ICP with local covariance modeling
- AA-ICP: Anderson Acceleration enhanced ICP for faster convergence
- NDT: Normal Distributions Transform for robust registration
Key features:
- CRTP-based design following the existing architecture
- Support for parallel optimization (configurable)
- Comprehensive unit tests and performance benchmarks
- Fixed convergence checking in base class to skip first iteration
- Added proper normal computation for Point-to-Plane ICP
- Adjustable parameters for each algorithm
Note: AA-ICP implementation has known issues and needs further debugging.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>1 parent 86758ad commit ae9521f
File tree
21 files changed
+4742
-90
lines changed- benchmark
- pcl
- example
- src/include/cpp-toolbox/pcl/registration
- impl
- test/pcl
21 files changed
+4742
-90
lines changed| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
12 | 12 | | |
13 | 13 | | |
14 | 14 | | |
| 15 | + | |
15 | 16 | | |
16 | 17 | | |
17 | 18 | | |
| |||
0 commit comments