Random Sample ICP algorithm for 3D point clouds. More...
#include <Icp.hpp>
Inheritance diagram for TRTK::RandomSampleIcp3D< ValueType >: Collaboration diagram for TRTK::RandomSampleIcp3D< ValueType >:Public Member Functions | |
| ValueType | compute () |
| Call this function to perform the transformation estimation. The RMSE is returned. | |
| void | setMaximumNumberIterationsFirstStage (unsigned value) |
| Sets the maximum number of iterations in the first stage (default is 20). | |
| void | setMaximumNumberIterationsSecondStage (unsigned value) |
| Sets the maximum number of iterations in the second stage (default is 20). | |
| void | setPercentage (unsigned value) |
| Sets the portion of randomly selected samples in the first stage. 0 means no samples and 100 means all samples (default is 50). | |
| void | setSourcePoints (const std::vector< Coordinate< ValueType > > &source_points) |
| Set the data points as decribed here. | |
| void | setTargetPoints (const std::vector< Coordinate< ValueType > > &target_points) |
| Set the model points as decribed here. | |
| void | setInitialEstimation (const Matrix &transformation) |
| Set the initial transformation from the source to the target point coordinate system. | |
| void | setEstimationAlgorithm (EstimateTransformation3D< ValueType > &estimator) |
| Set the transformation estimator. | |
| void | setMaximumNumberIterations (unsigned number) |
| Set the maximum number of iterations performed by an algorithm (default is 20). | |
| void | setMaximumRMSE (ValueType value) |
| Set an upper bound of the root mean square error; an algorithm terminates if the error of the current estimation falls below this value (default is 0). | |
| void | setEpsApproximate (float value) |
| Set the value for the eps-approximate neighbor search. The higher the value the faster the search (but also the less accurate) (default is 0). | |
| void | abortComputation () |
| Call this function to abort a running algorithm. | |
| const Matrix & | getTransformation () |
| Returns the estimated transformation in the form of a homogeneous matrix. | |
Public Attributes | |
| Signal< unsigned > | progress |
| Provides the current state of computation (from 0 to 100). | |
Protected Attributes | |
| unsigned | dimension |
| Dimension of the source and target points. | |
|
const std::vector< Coordinate < ValueType > > * | source_points |
| Reference to the data points given by the user. An algorithm might store a local copy which is transformed during the registration procedure. | |
| flann::Matrix< ValueType > | target_points |
| Internal copy of the model points given by the user. | |
| flann::Matrix< ValueType > | query_point |
| Query point used to search the kd-tree. | |
| flann::Matrix< ValueType > | distances |
| Distance to the nearest neighbor obtained by a nearest neighbor search. | |
| flann::Matrix< int > | indices |
| Index of the nearest neighbor in target_points obtained by a nearest neighbor search. | |
|
flann::Index< flann::L2 < ValueType > > * | tree |
| Data structure of the search index (kd-tree). | |
| float | eps |
| Eps-approximate value for the nearest neighbor search. | |
|
EstimateTransformation3D < ValueType > * | estimator |
| Reference to the user-provided estimator. | |
| unsigned | maximum_number_iterations |
| Maximum possible number of iterations during the computation. | |
| ValueType | maximum_rms_error |
| Iteration stop criterion. If the current RMSE falls below this bound, an algorithm might terminate. | |
| Matrix | transformation |
| Internally stored (homogeneous) transformation. | |
| volatile bool | abort |
| This flag must be checked during iterations; can be set by other threads through abortComputation(). | |
Random Sample ICP algorithm for 3D point clouds.
| ValueType | Scalar type (must be a floating point type). |
This algorithm implements a variant of the iterative closest point algorithm.
Basically, it proceeds in two stages. First, the classical iterative closest point algorithm is computed from a subset of data points. This subset is newly created in each iteration by randomly selecting points from the data points set. Since the subset is constantly changing, the convergence is not monotonicly decreasing anymore and the root mean square error might become greater. Hence, after each iteration the current estimate is saved if it is better than a previous estimate (measured by the RMSE). Second, the classical iterative closest point algorithm is run where its intial estimte is set to the best estimate from the first stage.
The maximum number of iterations in the first stage can be set by setMaximumNumberIterationsFirstStage(). The maximum number of iterations in the second stage by setMaximumNumberIterationsSecondStage() or by setMaximumNumberIterations(). The portion of randomly selected samples in the first stage can be set by setPercentage().
For some additional information and an example of how to use this class, please have a look at the interface class.
Fieten et al., "Fast and Accurate Registration of Cranial CT Images With A-mode Ultrasound", International Journal of Computer Assisted Radiology and Surgery, 2009
Definition at line 985 of file Icp.hpp.
| ValueType TRTK::RandomSampleIcp3D< ValueType >::compute | ( | ) | [virtual] |
Call this function to perform the transformation estimation. The RMSE is returned.
| range_error | A range error exception is thrown if no model points are given. |
Implements TRTK::IcpBase< ValueType >.
| void TRTK::IcpBase< ValueType >::setSourcePoints | ( | const std::vector< Coordinate< ValueType > > & | source_points ) | [virtual, inherited] |
Set the data points as decribed here.
Only a reference is stored.
Implements TRTK::IcpInterface< ValueType >.
| void TRTK::IcpBase< ValueType >::setTargetPoints | ( | const std::vector< Coordinate< ValueType > > & | target_points ) | [virtual, inherited] |
Set the model points as decribed here.
The targe points are copied.
Implements TRTK::IcpInterface< ValueType >.
Documentation generated by Doxygen
