This project proposes methods for navigation of an unmanned aerial vehicle (UAV) utilizing computer vision. This is achieved using ArUco marker boards which are used for pose estimation of the UAV. The goal is to have a UAV to fly autonomously using GPS in outdoor environments from where missions can be executed. When the UAV needs to recharge, a reliable GPS to vision transition is performed where the UAV uses its front camera to detect an ArUco marker board located on a wall to the entrance of a building. When the UAV is of a certain distance in front of this board located on the wall, it will start using its bottom camera for navigation in indoor GPS denied environments. In the indoor environment, a precise vision based landing can then be performed when it needs to recharge, by using its front camera for pose estimation of ArUco marker boards located in front of the landing stations. For vision based navigation, sensor fusion will be used for pose optimizations. An illustration of this can be seen in the video below. The master thesis for this project can be accessed using report.
To achieve autonomous flight for offboard control, the robot operating system (ROS) is used along with the PX4 autopilot. Using ROS, a number of nodes for drone control, autonomous flight, ArUco marker detection and sensor fusion can be run concurrently. The ArUco marker detection and pose estimation are accomplished by using OpenCV and an unscented Kalman filter for sensor fusion. The implementation has been tested in simulation using Gazebo and real flight using a Holybro Pixhawk 4 mini QAV250 Quadcopter with a Raspberry Pi 4B which was used as companion computer where an OptiTrack system has been used for ground truth tracking of the UAV.
Results show that the implementation works well, where the UAV was able to make GPS to vision transitions even in strong wind as well as its ability to use a minimum amount of ArUco markers located on the ground for vision based navigation. The latter was accomplished by successfully implementation of sensor fusion where IMU, barometer and vision data were fused together to achieve reliable pose estimates. The mean error of the precision landings were found to be in the order of centimeters which concludes a reliable implementation of the vision based landings.
This project has been implemented on an Ubuntu 18.04 machine using the ROS, Gazebo and OpenCV. It relies on the following software which must be installed to make it work properly.
You can either download the software or clone it directly from your terminal. To use the latter you must install Git using the command line:
#Install Git
sudo apt install git
#Clone software
git clone https://github.com/Kenil16/master_project.gitNow the project can be installed with all dependencies by running the script init_setup.bash. Now change directory and run the initialization script:
#Change directory and execute the initialization script
cd software/ros_workspace/
. ./init_setup.bashNow the installation and setup of the system are completed. If one wants to install the necessary software one-by-one, the following steps must be completed.
Install PX4 dependencies:
#Install a packages via apt
sudo apt install astyle build-essential ccache clang clang-tidy cmake cppcheck doxygen file g++ gcc gdb git lcov make ninja-build python3 python3-dev python3-pip python3-setuptools python3-wheel rsync shellcheck unzip xsltproc zip libeigen3-dev libopencv-dev libroscpp-dev protobuf-compiler python-pip python3-pip ninja-build gstreamer1.0-plugins-bad gstreamer1.0-plugins-base gstreamer10-plugins-good gstreamer1.0-plugins-ugly libgstreamer-plugins-base1.0-dev libgstrtspserver-1.0-dev xvfb python-rosdep2
#Install packages via pip
pip install --user argparse cerberus empy jinja2 numpy packaging pandas psutil pygments pyros-genmsg pyserial pyulog pyyaml setuptools six toml wheel rosdep
#Install packages via pip3
pip3 install --user --upgrade empy jinja2 numpy packaging pyros-genmsg toml pyyaml pymavlinkInstall texmaker and miktex for plots in latex format and report reading
#Install texmaker
sudo apt install texmaker
#Register GPG key
sudo apt-key adv --keyserver hkp:https://keyserver.ubuntu.com:80 --recv-keys D6BC243565B2087BC3F897C9277A7293F59E4889
#Register installation source
echo "deb https://miktex.org/download/ubuntu bionic universe" | sudo tee /etc/apt/sources.list.d/miktex.list
#Install MiKTeX
sudo apt-get update
sudo apt-get install miktexROS melodic is used in this project and can be installed together with Gazebo:
#Init workspace
sudo rosdep init
sudo rosdep update
#Setup your sources.list
$ sudo sh -c 'echo "deb https://packages.ros.org/ros/ubuntu $(lsb_release -sc) main" > /etc/apt/sources.list.d/ros-latest.list'
#Set up your keys
$ sudo apt-key adv --keyserver 'hkp:https://keyserver.ubuntu.com:80' --recv-key C1CF6E31E6BADE8868B172B4F42ED6FBAB17C654
#Installation
$ sudo apt update
$ sudo apt install ros-melodic-desktop-fullMavros can be installed using:
#Install mavros along with extra packages
sudo apt install ros-melodic-mavros ros-melodic-mavros-extras -yInstall dependencies to run software scripts:
sudo apt update
sudo apt install ros-melodic-hector-gazebo-plugins dvipng
pip install pathlib numpy pandas opencv-contrib-python scipy pymavlink matplotlibUpdate Gazebo version (tested on 9.16.0)
#Setup your computer to accept software from packages.osrfoundation.org.
sudo sh -c 'echo "deb https://packages.osrfoundation.org/gazebo/ubuntu-stable `lsb_release -cs` main" > /etc/apt/sources.list.d/gazebo-stable.list'
#Setup keys
wget https://packages.osrfoundation.org/gazebo.key -O - | sudo apt-key add -
#Update version
sudo apt-get update
sudo apt-get install gazebo9
#Otherwise gazebo will not run
sudo apt upgrade libignition-math2Install GeographicLib datasets:
#Fetch the sofware, change permissions and execute script
wget https://raw.githubusercontent.com/mavlink/mavros/master/mavros/scripts/install_geographiclib_datasets.sh
sudo chmod 755 install_geographiclib_datasets.sh
sudo ./install_geographiclib_datasets.shPX4 must be installed and this setup assumes PX4 SITL to be installed from the /home directory. Another installation diretcory can easily be made, but then the symbolics links must be changed accordingly:
#Initialize PX4 SITL environment from your home folder
cd ~
mkdir PX4_SITL
cd PX4_SITL
git clone https://github.com/PX4/Firmware.git --branch v1.11.0
cd Firmware
git submodule update --init --recursive
#Make symlink to user software models to PX4 SITL
cd software/ros_workspace/
ln -s $PWD/PX4-software/init.d-posix/* /home/$USER/PX4_SITL/Firmware/ROMFS/px4fmu_common/init.d-posix/
ln -s $PWD/PX4-software/mixers/* /home/$USER/PX4_SITL/Firmware/ROMFS/px4fmu_common/mixers/
ln -s $PWD/PX4-software/models/* /home/$USER/PX4_SITL/Firmware/Tools/sitl_gazebo/models/
ln -s $PWD/PX4-software/worlds/* /home/$USER/PX4_SITL/Firmware/Tools/sitl_gazebo/worlds/
ln -s $PWD/PX4-software/launch/* /home/$USER/PX4_SITL/Firmware/launch/
#Build the PX4 SITL firmware
cd /home/$USER/PX4_SITL/Firmware
DONT_RUN=1 make px4_sitl_default gazeboTo setup the system you have to run the following configuration file from your terminal to link PX4 to the ROS environment:
#Be in your ROS workspace and run the setup script
cd software/ros_workspace/
. ./setup_gazebo.bashTo run the program you have to execute the following command from your terminal:
#Run roslaunch which starts all the nodes in the system
roslaunch px4 gazebo_sim_v1.0.launch worlds:=optitrack_big_board_onepattern_missing_markers.world drone_control_args:="idle" x:=-21.0 y:=0Here you can define the world from a number of possible setups. The UAV will be in idle state by default, but that can be changed to some mission if prefered.
-
Hold pose using ArUco pose estimation for GPS2Vision board witn no wind
-
Hold pose using ArUco pose estimation for GPS2Vision board witn 5-7 m/s wind
-
Hold pose using ArUco pose estimation for landing station one
-
Hold pose using ArUco pose estimation for landing station two
-
Hold pose using ArUco pose estimation for landing station three
-
Vision navigation using full ArUco board with 1.0 m/s horizontal velocity
-
Vision navigation using ArUco board with missing markers with 5.0 m/s horizontal velocity