reusex::geometry::detail Namespace Reference

Internals of detect_loop_edges, exposed ONLY so the geometry stages of the pipeline can be regression-tested without a database and without RGB-D fixtures (docs/STANDARDS.md §7; same rationale as plane_factor_consistent_residual() in PlaneGraphOptimizer.cpp). More...

Classes
struct	PanoResection
	One panorama's independent resection against one frame. More...

Functions
std::vector< int >	ransac_rigid (const Eigen::Matrix3Xd &src, const Eigen::Matrix3Xd &dst, const LoopClosureOptions &opt, std::mt19937 &rng, Eigen::Matrix4d &best_T)
	RANSAC 3D-3D rigid fit: estimate best_T with dst ~= best_T * src from putative correspondences (both 3xN, column k of src matching column k of dst).
std::vector< LoopEdge >	pcm_filter (std::vector< LoopEdge > edges, const std::vector< Eigen::Matrix4d > &seed, const LoopClosureOptions &opt)
	Pairwise Consistency Maximization (Mangelson et al.
std::vector< LoopEdge >	edges_from_resections (std::vector< PanoResection > resections, const std::vector< Eigen::Matrix4d > &seed_poses, const PanoramaLoopOptions &opt, const LoopClosureOptions &gates, PanoramaLoopResult *stats=nullptr)
	Turn one panorama's per-frame resections into gated loop edges.

Detailed Description

Internals of detect_loop_edges, exposed ONLY so the geometry stages of the pipeline can be regression-tested without a database and without RGB-D fixtures (docs/STANDARDS.md §7; same rationale as plane_factor_consistent_residual() in PlaneGraphOptimizer.cpp).

Internals exposed ONLY so the geometry and gating stages can be regression- tested without a database and without RGB-D fixtures (docs/STANDARDS.md §7; same rationale as LoopClosure's detail namespace).

Not part of the stable API — do not call from application code.

Not a stable API.

Function Documentation

edges_from_resections()

std::vector< LoopEdge > reusex::geometry::detail::edges_from_resections	(	std::vector< PanoResection >	resections,
		const std::vector< Eigen::Matrix4d > &	seed_poses,
		const PanoramaLoopOptions &	opt,
		const LoopClosureOptions &	gates,
		PanoramaLoopResult *	stats = nullptr )

Turn one panorama's per-frame resections into gated loop edges.

For each pair (A, B) with A < B by frame index: T_AB = T_pano_A^-1 · T_pano_B, i.e. pose(A)^-1 · pose(B).

Applies, in order: min_frame_gap, the seed-disagreement gates, and the per-panorama cap (highest joint support first, ties broken on (i, j) so the output is deterministic). Sigmas scale with joint inlier support.

stats is accumulated into when non-null (counters are added, not reset), so a caller can sum across panoramas.

pcm_filter()

std::vector< LoopEdge > reusex::geometry::detail::pcm_filter	(	std::vector< LoopEdge >	edges,
		const std::vector< Eigen::Matrix4d > &	seed,
		const LoopClosureOptions &	opt )

Pairwise Consistency Maximization (Mangelson et al.

2018): return the largest mutually consistent subset of edges, where two edges are consistent when chaining them through the seed odometry closes a cycle within opt.pcm_trans_threshold / opt.pcm_rot_threshold. Sets of two or fewer edges are returned unchanged (a pair cannot outvote itself).

ransac_rigid()

std::vector< int > reusex::geometry::detail::ransac_rigid	(	const Eigen::Matrix3Xd &	src,
		const Eigen::Matrix3Xd &	dst,
		const LoopClosureOptions &	opt,
		std::mt19937 &	rng,
		Eigen::Matrix4d &	best_T )

RANSAC 3D-3D rigid fit: estimate best_T with dst ~= best_T * src from putative correspondences (both 3xN, column k of src matching column k of dst).

Uses opt.ransac_iterations hypotheses and opt.ransac_inlier_dist as the inlier distance, refits on the full inlier set, and returns the inlier column indices (empty when fewer than 3 correspondences are given or no hypothesis was finite).