Install OpenCV 4 on Jetson TX2

OpenCV on TX2.md

OpenCV version: 4.3.0 Python version: 3.6

Prerequisite

Required libraries

sudo apt-get update
sudo apt-get install curl libssl-dev
sudo apt-get install ffmpeg

#From opencv documentation
sudo apt-get install build-essential
sudo apt-get install cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev
sudo apt-get install python-dev python3-dev libtbb2 libtbb-dev libjpeg-dev libpng-dev libtiff-dev libjasper-dev libdc1394-22-dev

#I recommand install numpy via pip
python -m pip install numpy

#For scipy pip installation
sudo apt-get install python-dev python3-dev
python -m pip install scipy

For numerical calculation acceleration support

sudo apt-get install libopenblas-base libopenblas-dev libatlas-base-dev libatlas3-base liblapack-dev liblapack3 liblapacke-dev

Openblas If you're not on aarch64, please use sudo find / -name openblas to find your openblas and change accordingly* Open opencv/cmake/OpenCVFindOpenBLAS.cmake, find SET(Open_BLAS_INCLUDE_SEARCH_PATHS, add

/usr/include/aarch64-linux-gnu
/usr/include/aarch64-linux-gnu/openblas

find SET(Open_BLAS_LIB_SEARCH_PATHS, add

/usr/lib/aarch64-linux-gnu
/usr/lib/aarch64-linux-gnu/openblas

Atlas Open opencv/cmake/OpenCVFindAtlas.cmake, find set(Atlas_INCLUDE_SEARCH_PATHS and set(Atlas_LIB_SEARCH_PATHS, add /usr/include/aarch64-linux-gnu/atlas /usr/lib/aarch64-linux-gnu/atlas accordingly.

Lapack will be automatically found if you configured openblas and atlas correctly (because in ubuntu distribution, openblas contains lapack header)

Nvidia video codec sdk

First, install nvidia video codec sdk: https://developer.nvidia.com/nvidia-video-codec-sdk/download Copy the library files to /usr/local/cuda/targets/aarch64-linux/lib, header files to /usr/local/cuda/targets/aarch64-linux/include.

Commands

Before installing, download opencv and opencv-contrib from github, and put opencv-contrib inside opencv folder

cd opencv-4.3.0
mkdir build && cd build
cmake -D CMAKE_BUILD_TYPE=Release \
      -D WITH_TBB=ON \
      -D WITH_CUDA=ON \
      -D ENABLE_FAST_MATH=ON \
      -D CUDA_FAST_MATH=ON \
      -D WITH_NVCUVID=ON \
      -D WITH_CUBLAS=ON \
      -D WITH_V4L=ON \
      -D WITH_OPENGL=ON \
      -D WITH_QT=ON \
      -D WITH_GSTREAMER=ON \
      -D BUILD_opencv_cudacodec=ON \
      -D OPENCV_ENABLE_NONFREE=ON \
      -D OPENCV_GENERATE_PKGCONFIG=ON \
      -D OPENCV_EXTRA_MODULES_PATH=../../opencv_contrib-4.3.0/modules/ \
      -D PYTHON_DEFAULT_EXECUTABLE=$(which python3) \
      -D ENABLE_CXX11=ON \
      ..
sudo make -j7
sudo make install

Errors

fatal error: Eigen/Core: No such file or directory

This error happens in opencv-4.1.1, in 4.3.0 it works fine.

In /home/nvidia/opencv-4.1.1/modules/core/include/opencv2/core/private.hpp, edit the line #include <Eigen/Core> into #include <eigen3/Eigen/Core>

No rule to make target /usr/lib/aarch64-linux-gnu/libGL.so

This error happens in opencv-4.1.1, in 4.3.0 it works fine.

Relink the file

sudo rm /usr/lib/aarch64-linux-gnu/libGL.so
sudo ln -s /usr/lib/aarch64-linux-gnu/libGL.so.1 /usr/lib/aarch64-linux-gnu/libGL.so

/usr/bin/ld: cannot find -lopencv_contrib when compiling some project after installing OpenCV

When you compile OpenCV, first make sure that -D OPENCV_GENERATE_PKGCONFIG=YES is included. To confirm it, run pkg-config --cflags --libs opencv and pkg-config --cflags --libs opencv4 and compare. The latter should definitely have normal result if you're using OpenCV 4 (the former will have normal result as well if you have installed any libopencv package from apt-get)

In your project's Makefile, there should be a line like

g++ -o a a.cpp `pkg-config --cflags --libs opencv`

use opencv4 instead of opencv

fatal error: ../src/cap_ffmpeg_api.hpp: No such file or directory

This error happens in opencv-4.1.1, in 4.3.0 it works fine.

change ../src/cap_ffmpeg_api.hpp with ../src/cap_ffmpeg_legacy_api.hpp