Move further towards KdTreeNanoflann

[mvieth]

Extend autoSelectMethod to types that do not have xyz-coordinate (i.e. features types)

[Copilot]

Pull request overview

This PR extends pcl::search::autoSelectMethod usage beyond XYZ point types (notably toward feature/descriptor types) and updates several tools and registration components to rely on the auto-selected search backend, while also introducing compile-time dimension constants (NR_DIMS) in point representations to support KdTreeNanoflann specialization.

Changes:

• Replace direct KdTree/KdTreeFLANN usage with pcl::search::autoSelectMethod(...) in tools and registration feature search paths.
• Broaden precompiled instantiations of AutoSelectMethod and update autoSelectMethod internals to avoid xyz-only code paths for non-xyz types.
• Add NR_DIMS compile-time constants to point/feature representations and add a unit test to assert consistency with runtime dimensions.

Reviewed changes

Copilot reviewed 15 out of 15 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tools/pcd_viewer.cpp	Switch point picking search to autoSelectMethod and Search::Ptr.
tools/compute_hausdorff.cpp	Use autoSelectMethod for 1-NN distance computations.
tools/boundary_estimation.cpp	Remove unused commented search-method line.
test/common/test_common.cpp	Add coverage asserting DefaultPointRepresentation::NR_DIMS matches runtime dimensions.
search/src/auto.cpp	Instantiate AutoSelectMethod for PCL_POINT_TYPES (incl. feature types).
search/include/pcl/search/impl/auto.hpp	Extend selection logic for non-xyz types and specialize KdTreeNanoflann with NR_DIMS.
registration/include/pcl/registration/sample_consensus_prerejective.h	Change feature tree type to pcl::search::Search<FeatureT>::Ptr.
registration/include/pcl/registration/impl/sample_consensus_prerejective.hpp	Build feature search tree via autoSelectMethod.
registration/include/pcl/registration/ia_ransac.h	Change feature tree type to pcl::search::Search<FeatureT>::Ptr and adjust ctor init.
registration/include/pcl/registration/impl/ia_ransac.hpp	Build feature search tree via autoSelectMethod.
registration/include/pcl/registration/ia_fpcs.h	Replace reciprocal kd-tree alias with pcl::search::Search<PointSource>.
registration/include/pcl/registration/impl/ia_fpcs.hpp	Use autoSelectMethod for intermediate-point radius searches.
people/include/pcl/people/impl/ground_based_people_detection_app.hpp	Rely on EuclideanClusterExtraction’s internal auto-selected tree (remove manual tree wiring).
common/include/pcl/point_representation.h	Add NR_DIMS for default point/feature representations and propagate into specializations.
common/include/pcl/impl/point_types.hpp	Add descriptorSize trait specialization for NormalBasedSignature12.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[larshg]

Should we add it as an optional agurment to autoSelectMethod ?
Else there is no way to change it - as it can't be done afterwards.
One can still create an octree "manually" if one knows it better fits.

[mvieth]

I would rather not add the octree resolution as an argument. I think autoSelectMethod should hide any details related to the specific search method used, and determine all settings itself (e.g. resolution, or number of points per node for KdTreeNanoflann). I think having a resolution argument would kind of defeat the purpose of autoSelectMethod: being able to just request a search object without worrying which method is used or with which parameters. And like you said - if someone specifically wants an octree with a certain resolution, they can always create it manually and e.g. pass it to any PCL class via setSearchMethod.
I have added octree here only preliminarily because I think it will be faster than the BruteForce search even if the octree resolution is not perfectly chosen. I think in the future we could try to determine a good resolution via the 3D bounding box of the cloud, or we check whether we make the dynamic depth feature available in pcl::search::Octree.