3d-point-cloud: A repository from estambolieva

Installing Python Requirements

To install the packages we need to the lidar data exploration and ML training, it is only needed to install the requirements.txt file by typing in the terminal:

pip3 install -r requirements.txt

# Install PCLPY
conda install -c conda-forge -c davidcaron pclpy

# install PCL
sudo apt-get update -y
sudo apt-get install libpcl-dev -y

# conda install -c sirokujira pcl --channel conda-forge UNINSTALLED
# conda install -c sirokujira python-pcl --channel conda-forge WAS NEVER INSTALLED

Python-pcl works only with Python2.7, Python3.5 and Python3.6. I use Python3.8 so a virtual environment which runs python3.6 needs to be installed and used.

# Install Python3.6
sudo add-apt-repository ppa:deadsnakes/ppa
sudo apt-get update
sudo apt-get install python3.6
sudo apt autoremove

# Install virtual environment
sudo apt-get update
sudo apt-get install python3-virtualenv

# Create Python3.6 virtual environment called LIDAR
virtualenv -p /usr/bin/python3.6 lidar

# Work on this virtual environment
source lidar/bin/activate

# check the python version to verify that it is Python3.6
python --version

Error: when trying to import pcl in Python3.6, the following error occurs: ImportError: libpcl_keypoints.so.1.7: cannot open shared object file: No such file or directory

a. Tried Installing pcl v0.3.0rc1 from source works around this issue with libpcl_keypoints.so.1.7 on Ubuntu 18.04. Let's do this then

Download python-pcl-0.3.0a1.tar.gz from source - here
Unpack and install

tar zxf python-pcl-0.3.0a1.tar.gz
cd python-pcl-0.3.0a1
python setup.py install

> running install
> running egg_info
> writing python_pcl.egg-info/PKG-INFO
> writing dependency_links to python_pcl.egg-info/dependency_links.txt
> writing requirements to python_pcl.egg-info/requires.txt
> writing top-level names to python_pcl.egg-info/top_level.txt
> reading manifest file 'python_pcl.egg-info/SOURCES.txt'
> writing manifest file 'python_pcl.egg-info/SOURCES.txt'
> installing library code to build/bdist.linux-x86_64/egg
> running install_lib
> running build_py
> running build_ext
> cythoning pcl/_pcl_180.pyx to pcl/_pcl_180.cpp
> error: /home/path/python-pcl-0.3.0a1/pcl/_pcl_180.pyx

Tried Python-pcl works with libpcl_keypoints.so.1.7 however Ubuntu installs the latest version of it which currently is libpcl_keypoints.so.1.8. Installing libpcl_keypoints.so.1.7 downloaded as .deb from here

sudo apt remove libpcl-dev # to remove installed version 1.8

# checked the installed version
sudo apt list libpcl-dev

> libpcl-dev/bionic-updates 1.8.1+dfsg1-2ubuntu2.18.04.1 amd64
> N: There is 1 additional version. Please use the '-a' switch to see it

In order to add libpcl_keypoints.so.1.7 I needed to update the source.list of Ubuntu 18.04 ( Ubuntu 16.04 source.list ) and install I read this here:

sudo apt-get install libogdi-dev libhdf5-openmpi-dev python-mpi4py libgdal-dev libhdf5-mpi-dev python-vtk6 libhdf5-openmpi-100 libvtk6.3 libvtk6.3-qt libvtk6-qt-dev libvtk6-dev libboost-all-dev libogdi3.2 libopenmpi1.10 libgdal1i libgdal.so.1-1.11.3 libvtk6.2 libpcl-visualization1.7 libpcl-surface1.7 libpcl-io1.7 libpcl-apps1.7 libpcl1.7 libpcl-dev=1.7.2-14build1 --fix-broken

if these do not work, I need to reverse the changes!

Test that python-pcl has been installed successfully:

python3

>

https://vtk.org/download/ https://discourse.vtk.org/t/installing-vtk-in-ubuntu-18-04/2147/3

Viewing the Data

On Ubuntu

Using the displaz library.

To start displaz, type in the terminal the following independent of the folder in which you are right now:

displaz

Visualizing .pcd data

sudo apt-get install pcl-tools 

pcl_viewer multiview 1 ./path_to.pcd

Labelling the Data

Supervise.ly is an online tool that has been developed to label data for Machine Learning applications.

Luckily, they recently released their support for labelling 3D Point Cloud Data - link here. We will use this tool to label the data using their graphical interface to prepare it for the machine learning job. Here is a Medium blog post explaining how the labelling interface works.

It is needed to create an account in order to use Supervise.ly.

Supervise.ly allowed lidar data formats

Supervise.ly allows for 2 types of lidar data files to be uploaded to their platform: .pcd and .kitti. I have chosen to convert the .las file we have into a .pcd file.

Converting .las to .pcd

Note PDAL failed to correctly convert the files. The resulting .pcd file has 0 point cloud points and no header. Will try converting using open3d - which is a Python library.

Current Setup Operating System: Ubuntu 18.04

Needed Package PDAL

PDAL Installation Instructions here

 sudo apt-get update -y 
 sudo apt-get install -y pdal

File Convertion Instructions here

pdal translate input.las output.pcd

Result

When this is finished, please upload the .pcd file to Supervise.ly and start labelling.

Other Labelling Resources Explored

3D Lidar Annotator

Additional Resources

Installation Instructions

Link with these instructoins here.
PCL installation instructions here
PCLPY installation instructions here and here on Github

Python Packages

Las2Pcd Github Page here

sudo aptitude install libpcl-dev -y pip3 install python-pcl

Error: ImportError: libpcl_surface.so.1.7: cannot open shared object file: No such file or directory

sudo aptitude install libpcl-dev=1.7.2-14build1

download python-pcl tar.gz from there: https://github.com/strawlab/python-pcl/archive/v0.3.0rc1.tar.gz

image from GIMP

sudo aptitude install libpcl-dev=1.7.2-14build1
sudo apt install pcl-tools
sudo apt-get update -y 
sudo apt-get install -y cython
pip3 install cython
pip3 install numpy
sudo apt-get install libpq-dev python3-dev build-essential libxml2-dev libxslt1-dev libldap2-dev libsasl2-dev libffi-dev

>>>>>>>
>python3 setup.py install
>setup.py: error: cannot find PCL, tried
>    pkg-config pcl_common-1.8
>    pkg-config pcl_common-1.7
>    pkg-config pcl_common-1.6
>    pkg-config pcl_common


# check all libraries installed
pkg-config --list-all

https://github.com/PointCloudLibrary/pcl/archive/pcl-1.11.1.tar.gz & how to compile: https://pcl-tutorials.readthedocs.io/en/latest/compiling_pcl_posix.html

metslib

# install instructions here: https://kezunlin.me/post/137aa5fc/
wget https://www.coin-or.org/download/source/metslib/metslib-0.5.3.tgz
tar xzvf metslib-0.5.3.tgz
cd metslib-0.5.3
./configure
make
sudo make install

VTK

git clone https://gitlab.kitware.com/vtk/vtk.git
mkdir VTK-build # this is at the same level as the vtk git repo dir - not a child of it!!
cd VTK-build
cmake -DCMAKE_BUILD_TYPE:STRING=Release ../vtk/

# update cmake # download page: https://cmake.org/download/
wget https://github.com/Kitware/CMake/releases/download/v3.18.2/cmake-3.18.2.tar.gz
tar xvf cmake-3.18.2.tar.gz
cd cmake-3.18.2
./configure
make
sudo make install
sudo ln -s /usr/local/bin/cmake /usr/bin # so that cmake --version can find cname in /usr/bin/cmake

# get Q5 https://vtk.org/Wiki/VTK/Building/Linux
wget http://download.qt-project.org/official_releases/qt/5.15/5.15.1/single/qt-everywhere-src-5.15.1.tar.xz
tar -xf qt-everywhere-src-5.15.1.tar.xz
cd qt-everywhere-src-5.15.1
./configure
make -j2
sudo make install

FINISHED HERE... STILL TASKS TO DO...

sudo apt-get install libeigen3-dev
tar xvf pcl-pcl-1.11.1.tar.gz
cd pcl-pcl-1.11.1 && mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release -DBUILD_visualization=ON #cmake .. NEED TO INSTALL VTK FOR VISUALIZATION
# -- Could NOT find Pcap (missing: PCAP_LIBRARIES PCAP_INCLUDE_DIRS) 
#  Could not find a package configuration file provided by "VTK" with any of
#  the following names:

#   VTKConfig.cmake
#   vtk-config.cmake
make -j2
sudo make -j2 install

> successfully installed.

conda install -c sirokujira python-pcl --channel conda-forge

UnsatisfiableError: The following specifications were found to be incompatible with the existing python installation in your environment:

Specifications:

python-pcl -> python[version='>=2.7,<2.8.0a0|>=3.6,<3.7.0a0|>=3.5,<3.6.0a0']

Your python: python=3.8

estambolieva/3d-point-cloud