struct mp2p_icp::CovFeatureParameters

Overview

Parameters of CovFeatureEvaluator. They define the common cost function every registration result is evaluated with, so changing any of them changes the meaning of the features: they are part of the feature version.

#include <IcpCovFeatures.h>

struct CovFeatureParameters
{
    // fields

    double map_voxel = 0.10;
    double scan_voxel = 0.10;
    size_t max_scan_points = 4096;
    size_t probe_points = 1024;
    double max_assoc_dist = 1.0;
    double cauchy_scale = 0.05;
    size_t normal_k = 12;
    double normal_max_dist = 1.0;
    double max_condition = 1e9;
    uint64_t seed = 1;
    std::vector<double> probe_scales = {1.0, 4.0, 16.0};
    static constexpr size_t kNumProbeScales = 3;
};

Fields

double map_voxel = 0.10

Voxel size of the reference cloud [m].

double scan_voxel = 0.10

Voxel size of the query scan [m].

size_t max_scan_points = 4096

Query points kept after voxelization.

size_t probe_points = 1024

Prefix of the (shuffled) query used for probing.

double max_assoc_dist = 1.0

Nearest-neighbor association gate [m].

double cauchy_scale = 0.05

Scale of the Cauchy robust kernel [m].

size_t normal_k = 12

Neighbors of a map point for its normal.

double normal_max_dist = 1.0

Farthest neighbor allowed in a normal fit [m].

double max_condition = 1e9

Eigenvalue floor of H, relative to the largest.

uint64_t seed = 1

Seed of the query subsampling.

std::vector<double> probe_scales = {1.0, 4.0, 16.0}

Probing step sizes, in units of the analytic standard deviation along each whitened axis. Exactly kNumProbeScales of them.