open3d::t::pipelines::registration Namespace Reference

Data Structures
class	ICPConvergenceCriteria
	Class that defines the convergence criteria of ICP. More...
class	RegistrationResult
class	RobustKernel
class	TransformationEstimation
class	TransformationEstimationForColoredICP
class	TransformationEstimationPointToPlane
class	TransformationEstimationPointToPoint

Enumerations
enum	RobustKernelMethod {   RobustKernelMethod::L2Loss = 0, RobustKernelMethod::L1Loss = 1, RobustKernelMethod::HuberLoss = 2, RobustKernelMethod::CauchyLoss = 3,   RobustKernelMethod::GMLoss = 4, RobustKernelMethod::TukeyLoss = 5, RobustKernelMethod::GeneralizedLoss = 6 }
enum	TransformationEstimationType { TransformationEstimationType::Unspecified = 0, TransformationEstimationType::PointToPoint = 1, TransformationEstimationType::PointToPlane = 2, TransformationEstimationType::ColoredICP = 3 }

Functions
RegistrationResult	EvaluateRegistration (const geometry::PointCloud &source, const geometry::PointCloud &target, double max_correspondence_distance, const core::Tensor &transformation=core::Tensor::Eye(4, core::Float64, core::Device("CPU:0")))
	Function for evaluating registration between point clouds. More...
RegistrationResult	ICP (const geometry::PointCloud &source, const geometry::PointCloud &target, const double max_correspondence_distance, const core::Tensor &init_source_to_target=core::Tensor::Eye(4, core::Float64, core::Device("CPU:0")), const TransformationEstimation &estimation=TransformationEstimationPointToPoint(), const ICPConvergenceCriteria &criteria=ICPConvergenceCriteria(), const double voxel_size=-1.0, const bool save_loss_log=false)
	Functions for ICP registration. More...
RegistrationResult	MultiScaleICP (const geometry::PointCloud &source, const geometry::PointCloud &target, const std::vector< double > &voxel_sizes, const std::vector< ICPConvergenceCriteria > &criteria_list, const std::vector< double > &max_correspondence_distances, const core::Tensor &init_source_to_target=core::Tensor::Eye(4, core::Float64, core::Device("CPU:0")), const TransformationEstimation &estimation=TransformationEstimationPointToPoint(), const bool save_loss_log=false)
	Functions for Multi-Scale ICP registration. It will run ICP on different voxel level, from coarse to dense. The vector of ICPConvergenceCriteria(relative fitness, relative rmse, max_iterations) contains the stoping condition for each voxel level. The length of voxel_sizes vector, criteria vector, max_correspondence_distances vector must be same, and voxel_sizes must contain positive values in strictly decreasing order [Lower the voxel size, higher is the resolution]. Only the last value of the voxel_sizes vector can be {-1}, as it allows to run on the original scale without downsampling. More...
core::Tensor	GetInformationMatrix (const geometry::PointCloud &source, const geometry::PointCloud &target, const double max_correspondence_distance, const core::Tensor &transformation)
	Computes Information Matrix, from the transfromation between source and target pointcloud. It returns the Information Matrix of shape {6, 6}, of dtype Float64 on device CPU:0. More...

Enumeration Type Documentation

RobustKernelMethod

enum open3d::t::pipelines::registration::RobustKernelMethod

strong

Enumerator
L2Loss
L1Loss
HuberLoss
CauchyLoss
GMLoss
TukeyLoss
GeneralizedLoss

TransformationEstimationType

enum open3d::t::pipelines::registration::TransformationEstimationType

strong

Enumerator
Unspecified
PointToPoint
PointToPlane
ColoredICP

Function Documentation

EvaluateRegistration()

RegistrationResult open3d::t::pipelines::registration::EvaluateRegistration	(	const geometry::PointCloud &	source,
		const geometry::PointCloud &	target,
		double	max_correspondence_distance,
		const core::Tensor &	transformation = core::Tensor::Eye(4, core::Float64, core::Device("CPU:0"))
	)

Function for evaluating registration between point clouds.

Parameters

source	The source point cloud. (Float32 or Float64 type).
target	The target point cloud. (Float32 or Float64 type).
max_correspondence_distance	Maximum correspondence points-pair distance.
transformation	The 4x4 transformation matrix to transform source to target of dtype Float64 on CPU device.

GetInformationMatrix()

core::Tensor open3d::t::pipelines::registration::GetInformationMatrix	(	const geometry::PointCloud &	source,
		const geometry::PointCloud &	target,
		const double	max_correspondence_distance,
		const core::Tensor &	transformation
	)

Computes Information Matrix, from the transfromation between source and target pointcloud. It returns the Information Matrix of shape {6, 6}, of dtype Float64 on device CPU:0.

Parameters

source	The source point cloud. (Float32 or Float64 type).
target	The target point cloud. (Float32 or Float64 type).
max_correspondence_distance	Maximum correspondence points-pair distance.
transformation	The 4x4 transformation matrix to transform source to target.

ICP()

RegistrationResult open3d::t::pipelines::registration::ICP	(	const geometry::PointCloud &	source,
		const geometry::PointCloud &	target,
		const double	max_correspondence_distance,
		const core::Tensor &	init_source_to_target = core::Tensor::Eye(4, core::Float64, core::Device("CPU:0")),
		const TransformationEstimation &	estimation = TransformationEstimationPointToPoint(),
		const ICPConvergenceCriteria &	criteria = ICPConvergenceCriteria(),
		const double	voxel_size = -1.0,
		const bool	save_loss_log = false
	)

Functions for ICP registration.

Parameters

source	The source point cloud. (Float32 or Float64 type).
target	The target point cloud. (Float32 or Float64 type).
max_correspondence_distance	Maximum correspondence points-pair distance.
init_source_to_target	Initial transformation estimation of type Float64 on CPU.
estimation	Estimation method.
criteria	Convergence criteria.
voxel_size	The input pointclouds will be down-sampled to this voxel_size scale. If voxel_size < 0, original scale will be used. However it is highly recommended to down-sample the point-cloud for performance. By default origianl scale of the point-cloud will be used.
save_loss_log	When True, it saves the iteration-wise values of fitness, inlier_rmse, transformation, scale, iteration in a TensorMap loss_log_ in RegsitrationResult. Default: False.

MultiScaleICP()

RegistrationResult open3d::t::pipelines::registration::MultiScaleICP	(	const geometry::PointCloud &	source,
		const geometry::PointCloud &	target,
		const std::vector< double > &	voxel_sizes,
		const std::vector< ICPConvergenceCriteria > &	criteria_list,
		const std::vector< double > &	max_correspondence_distances,
		const core::Tensor &	init_source_to_target = core::Tensor::Eye(4, core::Float64, core::Device("CPU:0")),
		const TransformationEstimation &	estimation = TransformationEstimationPointToPoint(),
		const bool	save_loss_log = false
	)

Functions for Multi-Scale ICP registration. It will run ICP on different voxel level, from coarse to dense. The vector of ICPConvergenceCriteria(relative fitness, relative rmse, max_iterations) contains the stoping condition for each voxel level. The length of voxel_sizes vector, criteria vector, max_correspondence_distances vector must be same, and voxel_sizes must contain positive values in strictly decreasing order [Lower the voxel size, higher is the resolution]. Only the last value of the voxel_sizes vector can be {-1}, as it allows to run on the original scale without downsampling.

Parameters

source	The source point cloud. (Float32 or Float64 type).
target	The target point cloud. (Float32 or Float64 type).
voxel_sizes	VectorDouble of voxel scales of type double.
criteria_list	Vector of ICPConvergenceCriteria objects for each scale.
max_correspondence_distances	VectorDouble of maximum correspondence points-pair distances of type double, for each iteration. Must be of same length as voxel_sizes and criterias.
init_source_to_target	Initial transformation estimation of type Float64 on CPU.
estimation	Estimation method.
save_loss_log	When True, it saves the iteration-wise values of fitness, inlier_rmse, transformation, scale, iteration in a TensorMap loss_log_ in RegsitrationResult. Default: False.