Ramón Calvo's picture
Ramón Calvo
Robotics Engineer

Custom OpenCV 3 installation with ROS Kinetic

Published on 17 Jan 2021

Before you start reading, I made a .deb package of OpenCV 3.3.1 ready to work on Ubuntu 16 and ROS Kinetic. You can download it here. Please note that I do NOT actively maintain this package, and that compiling the package yourself as explained below is a better approach security-wise.

ROS Kinetic comes with an OpenCV package that only contains its core functionality. In order to use some advanced features (anything contained in opencv-contrib), you will have to build OpenCV from source and manually include any module you want from opencv-contrib. But the ROS Kinetic packages are already precompiled to work with a specific OpenCV version. In order for this to work, version 3.3.1 is mandatory (not even 3.3.0 will do it).

To do the source installation, first clone both opencv and opencv-contrib and change to the 3.3.1 branch:

$ git clone https://github.com/opencv/opencv -b 3.3.1 --depth=1
$ git clone https://github.com/opencv/opencv_contrib -b 3.3.1 --depth=1

The --depth=1 flag indicates git not to download all the commit history, but only the files of the specified branch. This will speed up the download. If you were to contribute to OpenCV, you will need to download the entire commit history. But since we are only going to compile it, this is fine.

Next, go to the opencv and generate the necessary build files.

$ mkdir opencv/build && cd opencv/build
$ cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr/local \
	    -DOPENCV_EXTRA_MODULES_PATH=/your/path/to/opencv_contrib ..

Note that you need to specify the path to your opencv_contrib folder in the -DOPENCV_EXTRA_MODULES_PATH parameter. If you want to only build a set of modules, you can activate them with the -DBUILD_opencv_<module> parameters.

To build everything, simply run:

$ make -j$(nproc)

Note that if you have enough CPU cores but not enough RAM, your system may crash due to insufficient memory. This happened to me with 8 cores and 7Gb of free RAM.

Once compiled you can generate a .deb package so the process of installing and especially uninstalling is greatly simplified. To do so, execute the script generated in the opencv/build folder:

sudo checkinstall

checkinstall will track all the changes that would take place if you executed make install and produce a .deb package accordingly. You will be prompted to fill information such as package name, maintainer name and email, …

To use the custom OpenCV library installation instead of the ROS one, you need to specify in every CMakeLists.txt of every package you want using the source installation of OpenCV by using the following code:

find_package(OpenCV 3.3.1 REQUIRED core aruco)

add_executable(your_node src/your_node.cc)
target_link_libraries(your_node ${catkin_LIBRARIES} ${OpenCV_LIBRARIES})

In my case, I wanted to use the aruco module, so I specified it in the find_package function.