gnd_mcc: Ground Segmentation Algorithm

In lidR: Airborne LiDAR Data Manipulation and Visualization for Forestry Applications

Ground Segmentation Algorithm

Description

This function is made to be used in classify_ground. It implements an algorithm for segmentation of ground points base on a Multiscale Curvature Classification. This method is a strict implementation of the MCC algorithm made by Evans and Hudak. (2007) (see references) that relies on the authors' original source code written and exposed to R via the the RMCC package.

Usage

mcc(s = 1.5, t = 0.3)

Arguments

s	numeric. Scale parameter. The optimal scale parameter is a function of 1) the scale of the objects (e.g., trees) on the ground, and 2) the sampling interval (post spacing) of the lidar data.
t	numeric. Curvature threshold

Details

There are two parameters that the user must define, the scale (s) parameter and the curvature threshold (t). The optimal scale parameter is a function of 1) the scale of the objects (e.g., trees) on the ground, and 2) the sampling interval (post spacing) of the lidar data. Current lidar sensors are capable of collecting high density data (e.g., 8 pulses/m2) that translate to a spatial sampling frequency (post spacing) of 0.35 m/pulse (1/sqrt(8 pulses/m2) = 0.35 m/pulse), which is small relative to the size of mature trees and will oversample larger trees (i.e., nominally multiple returns/tree). Sparser lidar data (e.g., 0.25 pulses/m2) translate to a post spacing of 2.0 m/pulse (1/sqrt(0.25 pulses/m2) = 2.0 m/pulse), which would undersample trees and fail to sample some smaller trees (i.e., nominally <1 return/tree).

Therefore, a bit of trial and error is warranted to determine the best scale and curvature parameters to use. Select a las tile containing a good variety of canopy and terrain conditions, as it's likely the parameters that work best there will be applicable to the rest of your project area tiles. As a starting point: if the scale (post spacing) of the lidar survey is 1.5 m, then try 1.5. Try varying it up or down by 0.5 m increments to see if it produces a more desirable digital terrain model (DTM) interpolated from the classified ground returns in the output file. Use units that match the units of the lidar data. For example, if your lidar data were delivered in units of feet with a post spacing of 3 ft, set the scale parameter to 3, then try varying it up or down by 1 ft increments and compare the resulting interpolated DTMs. If the trees are large, then it's likely that a scale parameter of 1 m (3 ft) will produce a cleaner DTM than a scale parameter of 0.3 m (1 ft), even if the pulse density is 0.3 m (1 ft). As for the curvature threshold, a good starting value to try might be 0.3 (if data are in meters; 1 if data are in feet), and then try varying this up or down by 0.1 m increments (if data are in meters; 0.3 if data are in feet).

References

Evans, Jeffrey S.; Hudak, Andrew T. 2007. A multiscale curvature algorithm for classifying discrete return LiDAR in forested environments. IEEE Transactions on Geoscience and Remote Sensing. 45(4): 1029-1038.

See Also

Other ground segmentation algorithms: gnd_csf, gnd_pmf

Examples

if (require(RMCC, quietly = TRUE))
{
LASfile <- system.file("extdata", "Topography.laz", package="lidR")
las <- readLAS(LASfile, select = "xyzrn", filter = "-inside 273450 5274350 273550 5274450")

las <- classify_ground(las, mcc(1.5,0.3))
#plot(las, color = "Classification")
}

lidR documentation built on Sept. 11, 2024, 5:21 p.m.