opecv.md

OpenCV V4 on Raspberry Pi 4, Buster

Why and How

• As of 9/2019, it's not possible to use pipenv install python3-opencv to install OpenCV v4.
• As of 9/2019, pipenv install opencv-contrib-python installs OpenCV v3.4.
• As of 10/2019, PiWheels still only has 3.4.3.18 listed for cp37m for opencv-contrib-python
  • See: https://www.piwheels.org/project/opencv-contrib-python/ to check status
• This is a modified set of instructions on how build OpenCV v4 from source, onRaspberry Pi 4, Buster
  • Based on: https://www.pyimagesearch.com/2018/09/26/install-opencv-4-on-your-raspberry-pi/
  • And for deps and compile flags: https://www.pyimagesearch.com/2019/09/16/install-opencv-4-on-raspberry-pi-4-and-raspbian-buster/

Prepare

• Update system libraries:
  • sudo apt-get update && sudo apt-get upgrade
• Install dev tools (and cmake):
  • sudo apt-get install build-essential cmake unzip pkg-config
• Install video libraries:
  • sudo apt-get install libjpeg-dev libtiff5-dev libjasper-dev libpng-dev
  • sudo apt-get install libavcodec-dev libavformat-dev libswscale-dev libv4l-dev
  • sudo apt-get install libxvidcore-dev libx264-dev
• Install GTK:
  • sudo apt-get install libfontconfig1-dev libcairo2-dev
  • sudo apt-get install libgdk-pixbuf2.0-dev libpango1.0-dev
  • sudo apt-get install libgtk2.0-dev libgtk-3-dev
• Install numerical optimizations for OpenCV:
  • sudo apt-get install libatlas-base-dev gfortran
• Install HDF5 and Qt GUIs:
  • sudo apt-get install libhdf5-dev libhdf5-serial-dev libhdf5-103
  • sudo apt-get install libqtgui4 libqtwebkit4 libqt4-test python3-pyqt5
• Install Python 3 development headers:
  • sudo apt-get install python3-dev
• TODO: Determine which of these libraries are strictly necessary for our deployment.

Create Virtual Environment

• This assumes you're using a python version manager (such as pyenv) and the current (global or local) python is set to Python 3.7.3
• Typically, work is done in a ~/work direction
  • mkdir ~/work
  • cd ~/work
• Use pipenv to create a new virtual environment
  • mkdir ~/work/cv
  • cd ~/work/cv
  • pipenv --python `which python`
• Activate the virtual environment
  • cd ~/work/cv
  • pipenv shell
• Install NumPy prerequisite
  • cd ~/work/cv
  • pipenv install numpy
• Install PiCamera API:
  • pipenv install "picamera[array]"

OpenCV Version 4.1.1

• Source downloaded/prepared via
  • Typically, work is done in a ~/work direction
    • cd ~/work
  • OpenCV
    • wget https://github.com/opencv/opencv/archive/4.1.1.tar.gz
    • rename to opencv-4.1.1.tar.gz
    • extract: tar xvfz opencv-4.1.1.tar.gz
    • mv opencv-4.1.1 opencv
  • OpenCV Contrib
    • wget https://github.com/opencv/opencv_contrib/archive/4.1.1.tar.gz
    • rename to opencv_contrib-4.1.1.tar.gz
    • extract: tar xvfz opencv_contrib-4.1.1.tar.gz
    • mv opencv_contrib-4.1.1 opencv_contrib

Prepare Build

• Work will be done in a build directory
  • mkdir ~/work/opencv/build
  • cd ~/work/opencv/build
• Run make command:

cmake -D CMAKE_BUILD_TYPE=RELEASE \
-D CMAKE_INSTALL_PREFIX=/usr/local \
-D OPENCV_EXTRA_MODULES_PATH=/home/pi/work/opencv_contrib/modules \
-D ENABLE_NEON=ON \
-D ENABLE_VFPV3=ON \
-D BUILD_TESTS=OFF \
-D INSTALL_PYTHON_EXAMPLES=OFF \
-D OPENCV_ENABLE_NONFREE=ON \
-D CMAKE_SHARED_LINKER_FLAGS=-latomic \
-D BUILD_EXAMPLES=OFF ..

• Note: OPENCV_EXTRA_MODULES_PATH is a full path to the opencv_contrib directory (plus /modules) that you created in a previous step.
• Note this disclaimer from OpenCV:

Update 2018-11-27: Notice the -D OPENCV_ENABLE_NONFREE=ON  flag. Setting this flag with OpenCV 4 ensures that you’ll have access to SIFT/SURF and other patented algorithms.

• Verify Python 3 interpreter and numpy library are coming from the pipenv virtual env location
  • e.g. for Python 3: /home/pi/.local/share/virtualenvs/cv-lRcGL5pf/bin/python

Backup Image File

• TODO: This may be a good time to backup the .img file to have a stored starting point.

Increase Swap File Size

• Edit the swap file config:
  • sudo nano /etc/dphys-swapfile
  • Change:

# CONF_SWAPSIZE=100
CONF_SWAPSIZE=2048

* Increases the swap file size to 2048MB
* The latest PyImageSearch instructions use a `CONF_SWAPSIZE` of 1024. But, I've kept it at 2048.

• Restart swap service:
  • sudo /etc/init.d/dphys-swapfile stop
  • sudo /etc/init.d/dphys-swapfile start

Build

• Run the make command across 4 cores:
  • make -j4
  • If this is unsuccessful, you can run make (wihthout -j4) to run single threaded. I was able to run this with multiple cores (with -j4)
• Install library:
  • sudo make install
  • sudo ldconfig

Undo Swap File Changes

• Set swap file back to its previous configuration settings
• Restart swap service

Link OpenCV Into Your Virtual Environment

• This step must be repeated for each Python virtual environment (which you create with pipenv) that you'd like to use import cv2
• Go to the virtual environment location:
  • e.g. cd ~/.local/share/virtualenvs/cv-lRcGL5pf/lib/python3.7/site-packages
  • link to the library:
    • e.g. ln -s /usr/local/lib/python3.7/site-packages/cv2/python-3.7/cv2.cpython-37m-arm-linux-gnueabihf.so cv2.so
    • The first argument after ln -s is the location of the OpenCV library which we just built. Make sure the path is correct.

Test the Install

• Go to your virtual environment, if you're not already there, and activate it
  • e.g. cd ~/work/cv
  • pipenv shell
  • Start a Python console: python
  • Import OpenCV: import cv2
    • There should be no errors, you should see the prompt (after a wait): >>>
  • Verify the OpenCV version: cv2.__version__
    • You should see the version we build: 4.1.1

Viewing GUI Via SSH

• On Mac, use XQuartz to view images via SSH connection
  • Open an XQuartz terminal
  • ssh -Y pivision2.local
  • export DISPLAY=":0"
  • Run OpenCV program via SSH connection