Crate modern_icp 

A modern modular pure Rust implementation of the Iterative Closest Point algorithm.

Example

// Basic usage
let (alignee_transform, error_sum) = Icp::new()
    .correspondence_estimator(NearestNeighbor::new(&target_cloud))
    .estimate_step_transform(point_to_plane_lls::estimate_isometry)
    .is_converged(same_squared_distance_error(0.1))
    .estimate_transform(alignee_cloud, &target_cloud);

// With outliers rejection and point filtering
let (alignee_transform, error_sum) = Icp::new()
    .correspondence_estimator(BidirectionalDistance::new(&target_cloud))
    .estimate_step_transform(point_to_plane_lls::estimate_isometry)
    .is_converged(never) // run for 20 iterations without convergence check
    .max_iterations(20) // stop after 20 iterations
    .filter_points(|pt: &PointCloudPoint<f32, 3>| pt.pos.z > 0.0) // only use points above the xy-plane
    .reject_outliers(reject_n_sigma_dist(3.0))
    .estimate_transform(alignee_cloud, &target_cloud);

Integrations

Serde

A serde integration is provided so you can serialize and deserialize PointCloud and PointCloudPoint using the serde crate. Enable the serde feature to use it.

Modelz

An integrations with the modelz crate is provided so you can use Model3D with the estimate_transform function. This allows for easy loading of 3D models from disk and using them with the ICP algorithm. Enable the modelz feature to use it.

use modelz::Model3D;

let Ok(alignee) = Model3D::load("alignee.gltf") else { return; };
let Ok(target) = Model3D::load("target.stl") else { return; };

let (alignee_transform, error_sum) = Icp::new()
    .correspondence_estimator(NearestNeighbor::new(&target))
    .estimate_step_transform(point_to_plane_lls::estimate_isometry)
    .is_converged(same_squared_distance_error(0.1))
    .estimate_transform(alignee, &target);

Rerun

An integration with the rerun crate is provided so you can visualize the ICP process. With the rerun viewer installed simply enable the rerun Cargo feature. When you run the ICP algorithm you’ll see the alignment process visualized in real-time.

Modules

convergence

correspondence

filter_points

icp

pca

reject_outliers

transform_estimation

Structs

Icp

The ICP algorithm for aligning two point clouds.

MaskedPointCloud

Plane

Plane that is described by the equation normal.dot(point_on_plane.coords) - constant == 0

PointCloudIterator

PointCloudPoint

Traits

ToPointCloud

Functions

compute_centroid

Computes the centroid (geometric center) of the point cloud.

demean_into_matrix

Demeans the point cloud by subtracting the centroid from every point and returns the demeaned point cloud as a matrix.

golden_section_search

See the paper from Dong et al.

kd_tree_of_point_cloud

point_cloud_from_position_slice

sum_squared_distances

transform_point_cloud

Transforms every point in the point cloud using the given transform.

Type Aliases

Plane3

PointCloud