一、删除系统自带的Opencv

使用jetpack安装的jetson，自带了opencv，但是没有cuda加速的，输入opencv_version查看。

使用jtop查看，可以确认自带的opencv是没用cuda的。

卸载opencv

pip3 uninstall opencv-python
sudo apt purge libopencv*
sudo apt autoremove
sudo apt update

二、安装OpenCV with CUDA的依赖

1）安装后面编译需要用到的依赖库

sudo apt install -y build-essential checkinstall cmake pkg-config yasm git gfortran
sudo apt install -y libjpeg8-dev libjasper-dev libtiff5-dev libavcodec-dev libavformat-dev libswscale-dev libdc1394-22-dev libxine2-dev libv4l-dev
sudo apt install -y libgstreamer1.0-dev libgstreamer-plugins-base1.0-dev libgtk2.0-dev libtbb-dev libatlas-base-dev libfaac-dev libmp3lame-dev libtheora-dev libvorbis-dev libxvidcore-dev libopencore-amrnb-dev libopencore-amrwb-dev x264 v4l-utils
sudo apt install python-numpy libtbb2 libtbb-dev libjpeg-dev libpng-dev libtiff-dev  libdc1394-22-dev

2）安装libpng12-0

sudo add-apt-repository ppa:linuxuprising/libpng12
sudo apt update
sudo apt install libpng12-0

若不慎安装了libpng12-dev，会出现如下报错，原因是libpng12-dev与libpng12-0冲突，则需要卸载libpng12-dev，再次安装libpng12-0。

sudo apt-get remove libpng12-dev

3）若没有安装python，需要安装python

终端输入python若报错，则安装sudo apt-get install python-dev或者sudo apt-get install python3-dev，需要安装python3.8版本，默认安装的就是3.8版本，ubuntu20.04已经去掉了python2版本。

三、下载编译安装OpenCV with CUDA

3.1 下载OpenCV源码

下载OpenCV4.5.5源码

https://github.com/opencv/opencv_contrib/archive/refs/tags/4.5.5.zip
https://github.com/opencv/opencv/archive/refs/tags/4.5.5.zip

其它版本见
https://github.com/opencv/opencv_contrib/tags
https://github.com/opencv/opencv/releases

3.2 编译OpenCV

1）新建文件夹~/opencv，把opencv-4.5.5.zip和opencv_contrib-4.5.5.zip文件都解压在该文件夹下面。

mkdir ~/opencv
cd ~/opencv
mkdir build &cd build
cd opencv-4.5.5
mkdir build &cd build

2）编译命令修改

这里需要注意的点如下：

1.修改CUDA Arch BIN的版本，通过jtop查看CUDA Arch BIN的版本

2.这里-DOPENCV_EXTRA_MODULES_PATH=/home/nvidia/opencv/opencv_contrib-4.5.5/modules \需要改为opencv_contrib-4.5.5.zip文件的路径

3）编译OpenCV

注意：如果有安装anaconda，建议关闭conda环境，进入主机环境后再编译，因为conda屏蔽了某些主机环境的库，导致编译时链接到了conda的库而冲突报错。

进入build文件夹，CMake编译生成Makefile文件

cmake \
-DCMAKE_BUILD_TYPE=Release \
-DCMAKE_INSTALL_PREFIX=/usr/local \
-DOPENCV_ENABLE_NONFREE=1 \
-DBUILD_opencv_python2=1 \
-DBUILD_opencv_python3=1 \
-DWITH_FFMPEG=1 \
-DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda \
-DCUDA_ARCH_BIN=8.7 \
-DCUDA_ARCH_PTX=8.7 \
-DWITH_CUDA=1 \
-DENABLE_FAST_MATH=1 \
-DCUDA_FAST_MATH=1 \
-DWITH_CUBLAS=1 \
-DOPENCV_GENERATE_PKGCONFIG=1 \
-DOPENCV_EXTRA_MODULES_PATH=/home/nvidia/opencv/opencv_contrib-4.5.5/modules \
..

make编译和安装，编译大概需要40min。

make -j8
sudo make install

4）安装opencv的python包

cd ~/opencv/opencv-4.5.5/build/python_loader
sudo pip3 install .

参考：
https://zhuanlan.zhihu.com/p/678890471
https://blog.csdn.net/weixin_52303102/article/details/136846684
https://blog.csdn.net/qq_39422817/article/details/134611876
https://blog.csdn.net/weixin_45901764/article/details/113943439

【AGX ORIN】系列(7)—AGX Orin安装OpenCV with CUDA