Skip to content

Latest commit

 

History

History
138 lines (110 loc) · 4.19 KB

install-opencv.md

File metadata and controls

138 lines (110 loc) · 4.19 KB
Raw

Install OpenCV

We use OpenCV for real-time computer vision. Official download page: link

Install OpenCV (Ubuntu)

sudo apt install libopencv-dev

Install OpenCV 3 (With contrib and Python 3 support)

Adapted from this post.

  1. Upgrade any pre-installed packages:

    $ sudo apt update
$ sudo apt upgrade

  2. Install developer tools used to compile OpenCV 3.0:

    $ sudo apt install build-essential cmake git pkg-config

  3. Install libraries and packages used to read various image and video formats from disk:

    $ sudo apt install libjpeg8-dev libtiff-dev libjasper-dev libpng12-dev
$ sudo apt install libavcodec-dev libavformat-dev libswscale-dev libv4l-dev

  4. Install GTK so we can use OpenCV’s GUI features:

    $ sudo apt install libgtk2.0-dev

  5. Install packages that are used to optimize various functions inside OpenCV, such as matrix operations:

    $ sudo apt install libatlas-base-dev gfortran

  6. Install the Python 3.4+ headers and development files:

    $ sudo apt install python3.4-dev

  7. Install pip

  8. Install Numpy

    $ sudo -H pip install numpy

  9. Download OpenCV main repo

    $ mkdir -p repos
$ cd repos
$ git clone https://github.com/Itseez/opencv.git
$ cd opencv
$ git checkout 3.0.0

  10. Download OpenCV contrib repo

$ cd ..
$ git clone https://github.com/Itseez/opencv_contrib.git
$ cd opencv_contrib
$ git checkout 3.0.0
  1. Configure OpenCV main repo
$ cd ../opencv
$ mkdir build
$ cd build
$ cmake -D CMAKE_BUILD_TYPE=RELEASE \
         -D CMAKE_INSTALL_PREFIX=/usr/local \
         -D INSTALL_C_EXAMPLES=ON \
         -D INSTALL_PYTHON_EXAMPLES=ON \
         -D OPENCV_EXTRA_MODULES_PATH=~/opencv_contrib/modules \
         -D BUILD_EXAMPLES=ON ..

The last step can also be done by hand using ccmake.

  1. Compile and install (This will probably take a long time, use the -j flag to speed up things using multithreaded compilation)
make -j$(nproc)  # compile
sudo make install; sudo ldconfig; cd  # install and go home

Install OpenCV 3 (With contrib, Python 3 support and fix for SVM_load)

OpenCV 3 has no method to load the SVM unless the specific commit where this was fixed is used (see this and this for related info).

To install OpenCV 3 with support for SVM loading, follow the previous steps, executing the following commands after step 10:

$ cd ../opencv
$ git checkout 37cbcf024c2b4160648299887a7c28d7cf28b1d3
$ cd ../opencv_contrib
$ git checkout 5409d5ad560523c85c6796cc5a009347072d883c

Then, continue with the previous guide, executing steps 11 and 12.

Install OpenCV 3 (With Python 3 support, CUDA 8 support, no NVIDIA driver installed)

To install OpenCV 3 with Python 3 support and CUDA 8 support (but no NVIDIA driver installed), follow the steps 1-9 of Install OpenCV 3 (With contrib and Python 3 support). Then, perform these steps:

  1. Patch the 3.0.0 branch to fix the issue with removed GraphCut in CUDA 8

    $ git format-patch -1 10896129b39655e19e4e7c529153cb5c2191a1db
$ git am < 0001-GraphCut-deprecated-in-CUDA-7.5-and-removed-in-8.0.patch

  2. Configure OpenCV main repo with CUDA and libcuda.so path

    $ mkdir build
$ cd build
$ cmake -D CMAKE_BUILD_TYPE=RELEASE \
         -D CMAKE_INSTALL_PREFIX=/usr/local \
         -D INSTALL_C_EXAMPLES=ON \
         -D INSTALL_PYTHON_EXAMPLES=ON \
         -D BUILD_EXAMPLES=ON \
         -D WITH_CUDA=ON \
         -D BUILD_CUDA_STUBS=ON \
         -D CMAKE_LIBRARY_PATH=/usr/local/cuda/lib64/stubs..

    The last step can also be done by hand using ccmake.

  3. Compile and install (This will probably take a long time, use the -j flag to speed up things using multithreaded compilation)

make -j$(nproc)  # compile
sudo make install; sudo ldconfig; cd  # install and go home