CspStandSegmentation (version 0.1.2)

csp_cost_segmentation: Comparative Shortest Path with cost weighting tree segmentation

Description

Segments single trees from forest point clouds based on tree positions (xyz-coordinates) provided in the map argument.

Usage

csp_cost_segmentation(
  las,
  map,
  Voxel_size = 0.3,
  V_w = 0,
  L_w = 0,
  S_w = 0,
  N_cores = 1
)

Value

Returns a copy of the las point cloud with an additional field for the TreeID.

Arguments

las

A lidR LAS S4 object.

map

A data.frame, including the columns X, Y, Z, TreeID, with X and Y depicting the location of the trees. Can be generated using CspStandSegmentation::find_base_coordinates_raster

Voxel_size

The voxel size (3D resolution) for the routing graph to determine the nearest map location for every point in the point cloud.

V_w

verticality weight. Since trunks are vertical structures, routing through voxels with high verticality can be rated 'cheaper.' should be a number between 0 and 1 with 0 meaning no benefit for more vertical structures.

L_w

Linearity weight. Similar to V_w but for linearity, higher values indicate a malus for linear shapes (usually branches).

S_w

Spericity weight. Similar to V_w but for sphericity, higher values indicate a malus for spherical shapes (usually small branches and leaves).

N_cores

number of CPU cores used for parallel routing using the foreach package.

Author

Julian Frey <[e-mail]>

Details

The whole point cloud is voxelized in the given resolution and the center of gravity for the points inside is calculated as voxel coordinate. A graph is build, which connects the voxel coordinates based on a db-scan algorithm. The distances between the voxel coordinates is weighted based on geometric features computed for the points in the voxels. Distances along planar and/or vertical faces like stems are weighted shorter than distances through voxels with high sphericity like leaves and clusters of twigs. This avoids small individuals spreading into the upper canopy. For every voxel center, the weighted distance in the network is calculated to all tree locations from the map argument. The TreeID of the map argument with the shortest distance is assigned to the voxel. All points in the point cloud receive the TreeID from their parent voxel.

See Also

comparative_shortest_path

Examples

# read example data
# \donttest{
file = system.file("extdata", "beech.las", package="CspStandSegmentation")
las = lidR::readTLSLAS(file)

# Find tree positions as starting points for segmentation
map <- CspStandSegmentation::find_base_coordinates_raster(las)

# segment trees
segmented <- las |>
CspStandSegmentation::add_geometry() |>
CspStandSegmentation::csp_cost_segmentation(map, 1, S_w = 0.5)


# }