Class HashedVoxelPointCloud

Defined in File HashedVoxelPointCloud.h

Nested Relationships

Nested Types

Inheritance Relationships

Base Types

public mrpt::maps::CMetricMap
public mrpt::maps::NearestNeighborsCapable

Class Documentation

class HashedVoxelPointCloud : public mrpt::maps::CMetricMap, public mrpt::maps::NearestNeighborsCapable

HashedVoxelPointCloud: a pointcloud stored as a sparse set of cubic voxels, indexed by a hash map. Efficient for storing point clouds, decimating them, and running nearest nearest-neighbor search.

Compile-time parameters

using global_index3d_t = index3d_t<int32_t>

using global_plain_index_t = uint64_t: collapsed plain unique ID for global indices

static constexpr std::size_t SSO_MAX_POINTS_PER_VOXEL = 32: Size of the std::array for the small-size optimization container in each voxel, defining the maximum number of points that can be stored without heap allocation.

static constexpr std::size_t GLOBAL_ID_SUBVOXEL_BITCOUNT = 5

Data structures

using point_vector_t = std::array<mrpt::math::TPoint3Df, SSO_MAX_POINTS_PER_VOXEL>

using grids_map_t = tsl::robin_map<global_index3d_t, VoxelData, index3d_hash<int32_t>>

Indices and coordinates

inline global_index3d_t coordToGlobalIdx(const mrpt::math::TPoint3Df &pt) const

inline mrpt::math::TPoint3Df globalIdxToCoord(const global_index3d_t idx) const: returns the coordinate of the voxel “bottom” corner

static inline global_plain_index_t g2plain(const global_index3d_t &g, int subVoxelIndex = 0)

Basic API for construction and main parameters

HashedVoxelPointCloud(float voxel_size = 1.00f)

Constructor / default ctor.

Parameters:: voxel_size – Voxel size [meters]

~HashedVoxelPointCloud()

void setVoxelProperties(float voxel_size): Reset the main voxel parameters, and clears all current map contents

Data access API

inline VoxelData *voxelByGlobalIdxs(const global_index3d_t &idx, bool createIfNew)

returns the voxeldata by global index coordinates, creating it or not if not found depending on createIfNew. Returns nullptr if not found and createIfNew is false

Function defined in the header file so compilers can optimize for literals “createIfNew”

inline const VoxelData *voxelByGlobalIdxs(const global_index3d_t &idx) const

inline VoxelData *voxelByCoords(const mrpt::math::TPoint3Df &pt, bool createIfNew): Get a voxeldata by (x,y,z) coordinates, creating the voxel if needed.

inline const VoxelData *voxelByCoords(const mrpt::math::TPoint3Df &pt) const

void insertPoint(const mrpt::math::TPoint3Df &pt): Insert one point into the dual voxel map

inline const grids_map_t &voxels() const

mrpt::math::TBoundingBoxf boundingBox() const override: Computes the bounding box of all the points, or (0,0 ,0,0, 0,0) if there are no points. Results are cached unless the map is somehow modified to avoid repeated calculations.

void visitAllPoints(const std::function<void(const mrpt::math::TPoint3Df&)> &f) const

void visitAllVoxels(const std::function<void(const global_index3d_t&, const VoxelData&)> &f) const

bool saveToTextFile(const std::string &file) const: Save to a text file. Each line contains “X Y Z” point coordinates. Returns false if any error occured, true elsewere.

API of the NearestNeighborsCapable virtual interface

bool nn_has_indices_or_ids() const override

size_t nn_index_count() const override

bool nn_single_search(const mrpt::math::TPoint3Df &query, mrpt::math::TPoint3Df &result, float &out_dist_sqr, uint64_t &resultIndexOrID) const override

bool nn_single_search(const mrpt::math::TPoint2Df &query, mrpt::math::TPoint2Df &result, float &out_dist_sqr, uint64_t &resultIndexOrID) const override

void nn_multiple_search(const mrpt::math::TPoint3Df &query, const size_t N, std::vector<mrpt::math::TPoint3Df> &results, std::vector<float> &out_dists_sqr, std::vector<uint64_t> &resultIndicesOrIDs) const override

void nn_multiple_search(const mrpt::math::TPoint2Df &query, const size_t N, std::vector<mrpt::math::TPoint2Df> &results, std::vector<float> &out_dists_sqr, std::vector<uint64_t> &resultIndicesOrIDs) const override

void nn_radius_search(const mrpt::math::TPoint3Df &query, const float search_radius_sqr, std::vector<mrpt::math::TPoint3Df> &results, std::vector<float> &out_dists_sqr, std::vector<uint64_t> &resultIndicesOrIDs, size_t maxPoints) const override

void nn_radius_search(const mrpt::math::TPoint2Df &query, const float search_radius_sqr, std::vector<mrpt::math::TPoint2Df> &results, std::vector<float> &out_dists_sqr, std::vector<uint64_t> &resultIndicesOrIDs, size_t maxPoints) const override

template<size_t MAX_KNN> void nn_multiple_search_impl(const mrpt::math::TPoint3Df &query, const size_t N, std::vector<mrpt::math::TPoint3Df> &results, std::vector<float> &out_dists_sqr, std::vector<uint64_t> &resultIndicesOrIDs) const

Public virtual methods implementation for CMetricMap

std::string asString() const override: Returns a short description of the map.

void getVisualizationInto(mrpt::opengl::CSetOfObjects &outObj) const override

bool isEmpty() const override: Returns true if the map is empty

void saveMetricMapRepresentationToFile(const std::string &filNamePrefix) const override: This virtual method saves the map to a file “filNamePrefix”+< some_file_extension >, as an image or in any other applicable way (Notice that other methods to save the map may be implemented in classes implementing this virtual interface).

const mrpt::maps::CSimplePointsMap *getAsSimplePointsMap() const override: Returns a cached point cloud view of the hash map. Not efficient at all. Only for MOLA->ROS2 bridge.

Public Members

TInsertionOptions insertionOptions

TLikelihoodOptions likelihoodOptions

TRenderOptions renderOptions

float voxel_size = 1.00f

mola::HashedVoxelPointCloud::TInsertionOptions insertionOpts

mola::HashedVoxelPointCloud::TLikelihoodOptions likelihoodOpts

mola::HashedVoxelPointCloud::TRenderOptions renderOpts

Protected Functions

void internal_clear() override

struct TInsertionOptions : public mrpt::config::CLoadableOptions

Options for insertObservation()

Public Functions

TInsertionOptions() = default

void loadFromConfigFile(const mrpt::config::CConfigFileBase &source, const std::string &section) override

void dumpToTextStream(std::ostream &out) const override

void writeToStream(mrpt::serialization::CArchive &out) const

void readFromStream(mrpt::serialization::CArchive &in)

Public Members

uint32_t max_points_per_voxel = 0: Maximum number of points per voxel. 0 means no limit.

double remove_voxels_farther_than = .0: If !=0, remove the voxels farther (L1 distance) than this distance, in meters.

float min_distance_between_points = .0f: If !=0 skip the insertion of points that are closer than this distance to any other already in the voxel.

struct TLikelihoodOptions : public mrpt::config::CLoadableOptions

Options used when evaluating “computeObservationLikelihood” in the derived classes.