struct mp2p_icp::OptimalTF_Result¶
Overview¶
The is the output structure for all optimal transformation methods.
#include <OptimalTF_Result.h> struct OptimalTF_Result { // fields mrpt::poses::CPose3D optimalPose; double optimalScale = 1.0; OutlierIndices outliers; double gravity_information_share = -1.0; mrpt::math::CMatrixDouble66 measurement_information; };
Fields¶
OutlierIndices outliers
Correspondence that were detected as outliers.
double gravity_information_share = -1.0
Share of the final rotational information that came from the gravity prior, in [0,1], on the last Gauss-Newton iteration; <0 when no prior was given.
A verticality prior competes against the point pairs for the same two tilt DOFs, but it is expressed in radians while they are in metres, so its influence is set by the pair count and their lever arms and is not readable from sigma_rad alone. This reports what it actually was.
mrpt::math::CMatrixDouble66 measurement_information
Information (Gauss-Newton Hessian) of the pairings alone, on the last iteration: robust-kernel weights and the cov2cov scaling included, the pose and gravity priors excluded. In the SE(3) right tangent of optimalPose, translation first (the layout of OptimalTF_GN_Parameters::prior). All zeros if the method does not compute it.
Its inverse is the registration uncertainty the geometry supports. Kept apart from the priors, since a direction the pairings do not constrain (e.g. along a corridor) looks well constrained once a prior is added, and as information rather than covariance, which is ill-conditioned in exactly those directions.