B Appendix 2: Tree 3D

The aim of this appendix is to test tree segmentation and light on the las cloud from Dicorynia guyanensis in Paracou.

B.1 Tree Segmentation

Using lidr. Works well to delineate the tree cloud but not build back architecture. Can be a good step for AMAPvox.

Individual segmented with lidR.

Figure B.1: Individual segmented with lidR.

B.2 Tree architecture

The reconstruction of the architecture, i.e. rebuilding the trunk and the branches. On this point there are plenty of poorly documented, impossible to install and not so open tools. But after a long battle I managed to make treeseg work a bit despite its lack of real documentation. And as I suspected unfortunately the data doesn’t seem to be good enough for this task. Indeed, besides the fact that we are in a complex rainforest, this reconstruction is normally done from high resolution LiDAR TLS data. But here I have only few points that represent the trunk or the main branches. The algorithm is therefore unable to reconstruct the cylinders.

B.3 Tree light

For this part I am still waiting for Nicolas to share with me the trajectos (3D path of the drone) to use AMAPvox. Nevertheless I don’t see how this will orientate more the sampling than using a device on the spot like with the Sextonia, especially if we can’t reconstruct the architecture beforehand. So I think I’ll stop here and wait for the postdoc to face this calculation for the a posteriori analysis of the mutations.

B.4 Tree Segmentation

This v explored tree segmentation with QSM on the las cloud from Dicorynia guyanensis in Paracou.

B.4.1 treeseg

Failed, and no further documentation, dropping thus treeseg.

## Convert 
# las to pcd
pdal translate YS-20201123-130625classclipDycorinia.las test.pcd --writers.pcd.order="X=Float:2,Y=Float:2,Z=Float:2" # to avoid float issue
# visualize in cloud compare
tail -n +11 test.pcd > test.xyz
# computing bounding box
Rscript -e 'summary(readr::read_delim("test.xyz", col_names = F, delim = " ")[1:2])'
echo 286791 286811 583888 583910 > test.dat
# fixing variables
# open with geany and replace X Y Z by x y z (should be further automated)

## downsample 
# 1. resoluation in meters
~/Tools/treeseg/build/downsample 0.001 test.pcd # 0.04 online
~/Tools/treeseg/build/pcdPointTreeseg2txt test.tile.downsample.test.pcd

## getdemslice
# 1. the resolution of the DEM in meters
# 2) the percentile which is considered to constitute the ground (i.e., if all the points inside each DEM tile are lined up by their z coordinate - the point closest to the Nth percentile is considered the ground - sometimes useful for noisy data). 
# 3)&4) The zmin and zmax of the slice. So e.g.,:
~/Tools/treeseg/build/getdemslice 1 1 0 5 test.tile.downsample.test.pcd # 2 3 3 6
~/Tools/treeseg/build/pcdPointTreeseg2txt test.slice.pcd

## Find stems
~/Tools/treeseg/build/findstems 15 0.2 2 test.dat test.slice.pcd
~/Tools/treeseg/build/pcdPointTreeseg2txt test.intermediate.slice.clusters.pcd
~/Tools/treeseg/build/pcdPointTreeseg2txt test.intermediate.slice.clusters.regions.pcd

## Segments stems
~/Tools/treeseg/build/segmentstem 12.5 test.tile.downsample.test.pcd test.intermediate.slice.clusters.regions.pcd

B.4.2 Computree

B.4.2.1 Windows version

Not working with wine when calling downsampling with SimpleTree plugin. Test on voncyane windows validated the software, to be tested on Niklas computer thursday afternoon.

cd ~/Tools/Computree
wget http://rdinnovation.onf.fr/attachments/download/2788/SimpleForest_v5.1.3_computree_v5.0.221b.zip
unzip SimpleForest_v5.1.3_computree_v5.0.221b.zip 
cd SimpleForest_v5.1.3_computree_v5.0.221b/
wine CompuTreeGui.exe

Overview:

  1. Import the cloud
  2. Reduce points density
  3. Classify ground points (TLS)
  4. Ground point filter
  5. Statistical outlier filter
  6. Dtm pyramidal Mlesac fit
  7. Cut cloud above DTM
  8. Cut cloud above DTM
  9. Cut cloud above DTM
  10. Stem filter
  11. Statistical outlier filter
  12. Radius outlier filter
  13. Segmentation euclidean clustering
  14. Dikstra based tree segmentation
  15. Voroni based tree segmentation
  16. Height based tree identification from segmented clouds
  17. Height based tree identification from segmented clouds
  18. Cut cloud above dtm
  19. Cut cloud above dtm
  20. Stem filter
  21. Euclidean clustering filter
  22. Merge clouds from two groups
  23. Statistical outlier filter
  24. Euclidean clustering filter
  25. Sphere following
  26. QSM median filter
  27. QSM based tree clustering
  28. Sphere following advanced
  29. QSM median filter
  30. QSM correct shoots
  31. QSM allometric correction
  32. QSM to ground exploration
  33. QSM list exporter

B.4.2.2 Linux version

Still failing, waiting for Alexandre Piboule feedbacks.

cd ~/Tools/Computree
sudo apt-get update
sudo apt-get install subversion
sudo apt-get install build-essential qtcreator qt5-default qt5-doc qt5-doc-html qtbase5-doc-html qtbase5-examples
wget http://rdinnovation.onf.fr/attachments/download/2378/kit_dev_linux.zip
unzip kit_dev_linux.zip
cd kit_dev_linux
./recuperer_depots.sh
sudo apt-get install libopencv-dev 
sudo apt-get install libpcl-dev # pcl 1.10 need pcl 1.8
sudo apt-get install libgdal-dev
sudo apt-get install libgsl-dev
qtcreator
# > open > all.pro
# > projet > unactivate shadow build
# > projet/run > “Environnement d’exécution” > rajouter ” ;. ” à la fin de la variables LD_LIBRARY_PATH
qmake all.pro
# > pcl1.8 to pcl1.10 in code
sudo ln -s /usr/local/include/opencv4/opencv2/ /usr/local/include/opencv2 # fix opencv2/core/core.hpp issue
# > compile
# > run

B.4.3 Tree Segmentation with lidr

Works well to delineate the tree cloud but not to segment it. Can be a good step for AMAPvox.

require(lidR)
require(rlas) # Necessary for writelax
require(rgdal) # Writing to shp or raster
require(tictoc) # for tic() toc() function
require(rgl)
data <- "data/Paracou/AG/YS-20201123-130625classclipDycorinia.las"
las <- readLAS(data) 
# lascheck(las)
# summary(las)
# sort(unique(las@data$Classification))
# plot(las, color = "Classification")
las_class <- lasfilter(las, Classification == 1)
# plot(las_class)
dtm <- grid_terrain(las, algorithm = knnidw(k = 8, p = 2))
las_normalized <- lasnormalize(las, dtm)
lasfilternoise = function(las, sensitivity)
{
  p95 <- grid_metrics(las, ~quantile(Z, probs = 0.95), 10)
  las <- lasmergespatial(las, p95, "p95")
  las <- lasfilter(las, Z < p95*sensitivity)
  las$p95 <- NULL
  return(las)
}
las_denoised <- lasfilternoise(las_normalized, sensitivity = 1.2)
# plot(las_denoised)
# plot(las_normalized)
chm <- grid_canopy(las_denoised, 0.5, pitfree(c(0,2,5,10,15), c(3,1.5), subcircle = 0.2))
# plot_dtm3d(chm)
ker <- matrix(1,5,5)
chm_s <- focal(chm, w = ker, fun = median)
algo <- watershed(chm_s, th = 10, tol = 0.7, ext = 1)
las_watershed  <- lastrees(las_denoised, algo)
trees <- lasfilter(las_watershed, !is.na(treeID))
plot(trees, color = "treeID", colorPalette = pastel.colors(100))
path <- "/home/sylvain/Documents/ECOFOG/treemutation/data/Paracou/AG/YSsegmented.gif"
movie3d(spin3d(), duration = 5, movie = path)

B.4.4 Forest Metrics

Hard to install and no better than lidR as it does not segment the crown.

sudo apt-get update
sudo apt-get install git build-essential linux-libc-dev
sudo apt-get install cmake cmake-gui 
sudo apt-get install libeigen3-dev
sudo apt-get install libboost-all-dev
sudo apt-get install libflann-dev
sudo apt-get install libvtk6-qt-dev
sudo apt-get install libqhull-dev
sudo apt-get install libproj-dev 
sudo apt autoremove
# pcl skipping 1.10
wget https://github.com/PointCloudLibrary/pcl/archive/pcl-1.8.1.tar.gz
tar -xf pcl-1.8.1.tar.gz
cd pcl-pcl-1.8.1 && mkdir build && cd build
cmake ..
make
sudo make install
# ForestMetrics
git clone --recursive https://github.com/yurithefury/ForestMetrics.git ForestMetrics
cd ForestMetrics
mkdir build
cd build
cmake ..
make

B.4.5 sp4e

Segmenting cloud but no QSM.

cd ~/Tools
git clone https://github.com/mparkan/sp4e_project.git
cd sp4e_project

B.4.6 TreeQSM

Mathlab, giving this one up.