lidR-spatial-index: Spatial index

In Jean-Romain/lidR: Airborne LiDAR Data Manipulation and Visualization for Forestry Applications

Spatial index

Description

This document explains how to process point-clouds taking advantage of different spatial indexes available in the lidR package. lidR can use several types of spatial indexes to apply algorithms (that need a spatial indexing) as fast as possible. The choice of the spatial index depends on the type of point-cloud that is processed and the algorithm that is performed. lidR can use a grid partition, a voxel partition, a quadtree or an octree. See details.

Usage

sensor(las, h = FALSE)

sensor(las) <- value

index(las, h = FALSE)

index(las) <- value

Arguments

las	An object of class LAS or LAScatalog.
h	boolean. Human readable. Everything is stored as integers that are understood internally. Use h = TRUE for user readable output.
value	integer or character. A code for referring to a sensor type or a spatial index type. Use one of "unknown", "als", "tls", "uav", "dap", "multispectral" for sensor type and one of "auto", "gridpartition", "voxelpartition", "quadtree", "octree" for spatial index type.

Details

From lidR (>= 3.1.0), a LAS object records the sensor used to sample the point-cloud (ALS, TLS, UAV, DAP) as well as the spatial index that must be used for processing the point cloud. This can be set manually by the user but the simplest is to use one of the read*LAS() functions. By default a point-cloud is associated to a sensor and the best spatial index is chosen on-the-fly depending on the algorithm applied. It is possible to force the use of a specific spatial index.

Information relative to the spatial indexing is stored in slot ⁠@index⁠ that contains a list with two elements:

• sensor: an integer that records the sensor type

• index: an integer that records the spatial index to be used

By default the spatial index code is 0 ("automatic") meaning that each function is free to choose a different spatial index depending on the recorded sensor. If the code is not 0 then each function will be forced to used the spatial index that is imposed. This, obviously, applies only to functions that use spatial indexing.

LAScatalog objects also record such information that is automatically propagated to the LAS objects when processing.

Note: before version 3.1.0, point-clouds were all considered as ALS because lidR was originally designed for ALS. Consequently, for legacy and backwards-compatibility reasons, readLAS() and readALSLAS() are actually equivalent. readLAS() tags the point cloud with "unknown" sensor while readALSLAS() tags it with 'ALS'. Both behave the same and this is especially true compared with versions < 3.1. As a consequence, using readLAS() provides the same performance (no degradation) than in previous versions, while using one of the read*LAS() functions may improve the performance.

Examples

LASfile <- system.file("extdata", "example.laz", package="rlas")
las <- readLAS(LASfile)

# By default the sensor and spatial index codes are 0
sensor(las)
index(las)

# Codes are used internally and not intended to be known by users
# Use h option for human readable output
sensor(las, h = TRUE)
index(las, h = TRUE)

# Modification of the sensor enables users to select a better spatial index
# when processing the point-cloud.
sensor(las) <- "tls"
sensor(las, h = TRUE)
index(las, h = TRUE)

# Modification of the spatial index forces users to choose one of the available
# spatial indexes.
index(las) <- "quadtree"
sensor(las, h = TRUE)
index(las, h = TRUE)

# The simplest way to take advantage of appropriate spatial indexing is
# to use one of the read*LAS() functions.
las <- readTLSLAS(LASfile)
sensor(las, h = TRUE)
index(las, h = TRUE)

# But for some specific point-clouds / algorithms it might be advisable to force
# the use of a specific spatial index to perform the computation faster
index(las) <- "voxelpartition"
index(las, h = TRUE)

# With a LAScatalog, spatial indexing information is propagated to the
# different chunks
ctg = readTLSLAScatalog(LASfile)
index(ctg) <- "voxelpartition"
sensor(ctg, h = TRUE)
index(ctg, h = TRUE)

# ==================
# PERFORMANCE TESTS
# ==================

## Not run: 
# Performance tests on TLS
# ------------------------

# The package does not include TLS data
# so we can generate something that looks TLS-ish
# >>>>>>>>>>>
X <- runif(50, -25, 25)
Y <- runif(50, -25, 25)
X <- as.numeric(sapply(X, function(x) rnorm(2000, x, 2)))
Y <- as.numeric(sapply(Y, function(x) rnorm(2000, x, 2)))
Z <- abs(rnorm(length(Y), 10, 5))
veg <- data.frame(X,Y,Z)
X <- runif(5000, -30, 30)
Y <- runif(5000, -30, 30)
Z <- runif(5000, 0, 1)
ground <- data.frame(X,Y,Z)
X <- runif(30, -30, 30)
Y <- runif(30, -30, 30)
Z <- runif(30, 0, 30)
noise <- data.frame(X,Y,Z)
las <- LAS(rbind(ground, veg, noise))
# <<<<<<<<<<<<<

plot(las)

# If read with readALSLAS()
sensor(las) <- "als"
system.time(classify_noise(las, sor(20, 8)))
#> 1.5 sec

# If read with readTLSLAS()
sensor(las) <- "tls"
system.time(classify_noise(las, sor(20, 8)))
#> 0.6 sec

# Performance tests on ALS
# ------------------------

# The package does not include large ALS data
# so we can generate something that looks ALS-ish
# >>>>>>>>>>>
X <- runif(4e5, 0, 1000)
Y <- runif(4e5, 0, 1000)
Z <- 40*sin(0.01*X) + 50*cos(0.005*Y) + abs(rnorm(length(Y), 10, 5))
veg <- data.frame(X,Y,Z)
X <- runif(100, 0, 1000)
Y <- runif(100, 0, 1000)
Z <- 40*sin(0.01*X) + 50*cos(0.005*Y) + abs(rnorm(length(Y), 10, 5)) + runif(100, 30, 70)
noise <- data.frame(X,Y,Z)
las <- LAS(rbind(veg, noise))
# <<<<<<<<<<<<<

plot(las)

# If read with readALSLAS()
sensor(las) <- "als"
system.time(classify_noise(las, sor(15, 8)))
#> 3 sec

# If read with readTLSLAS()
sensor(las) <- "tls"
system.time(classify_noise(las, sor(15, 8)))
#> 4.3 sec

## End(Not run)

Jean-Romain/lidR documentation built on April 6, 2024, 9:41 p.m.