Module registration 

Registration algorithms

Structs

CloudScale

Where a cloud sits and how big it is: the centroid of its finite points and their RMS distance from it. ICP uses both to judge whether an update still moves the points by a meaningful amount.

ICPResult

Result of ICP registration

IcpScaleLevel

One level in a coarse-to-fine ICP pyramid.

LocalFrame

A frame centred on a point cloud, for running registration in f32.

MultiScaleIcpConfig

Coarse-to-fine point-to-point ICP configuration.

Functions

icp

ICP (Iterative Closest Point) registration - Main function matching requested API

icp_converged

Decide whether ICP has converged. Both tests are relative, so they behave the same on a small indoor scan as on a large outdoor one (an absolute change in MSE does not: MSE shrinks with the square of the scene size).

icp_detailed

Detailed ICP registration with comprehensive options and result

icp_legacyDeprecated

Legacy ICP function with different signature for backward compatibility

icp_point_to_plane

Point-to-plane ICP variant (requires target normals).

icp_point_to_plane_detailed

Detailed point-to-plane ICP with full parameter control.

icp_point_to_point

Point-to-point ICP registration

icp_point_to_point_default

Point-to-point ICP registration with default parameters

multiscale_icp_point_to_point

Coarse-to-fine point-to-point ICP using voxel downsampling at each scale.

rigid_transform_from_sums

Best rigid transform taking a set of source points onto their matched target points (Kabsch / SVD), from running sums over the matched pairs: the pair count, the sums of source and target points, and the sum of source * targetᵀ. Used by the CPU ICP and by the GPU ICP in threecrate-gpu, which computes the sums on the GPU.