ubuntu20.04 opencv 4.2 + opencv_contrib 4.2.0 安装笔记和 opencv 4.5.4 with cuda 安装笔记-526互联

参考：

https://docs.opencv.org/4.x/d7/d9f/tutorial_linux_install.html

Build with opencv_contrib

# 1. Install minimal prerequisites， libgtk2.0-dev pkg-config 用来显示图像
sudo apt update && sudo apt install -y cmake g++ wget unzip libgtk2.0-dev pkg-config

# 2. Download and unpack sources
# cd 到自己要放 opencv 的路径下，比如我们的是：/home/h/programs/cv
cd /home/h/programs/cv
sudo wget -O opencv.zip https://github.com/opencv/opencv/archive/4.2.0.zip --no-check-certificate
sudo wget -O opencv_contrib.zip https://github.com/opencv/opencv_contrib/archive/4.2.0.zip --no-check-certificate
unzip opencv.zip && unzip opencv_contrib.zip

# 3. Create build directory and switch into it
mkdir build && cd build

# 4. Configure， build 目录内, 我们把 opencv 安装到 $HOME/.local  下
cmake -DCMAKE_INSTALL_PREFIX=$HOME/.local  -DOPENCV_EXTRA_MODULES_PATH=../opencv_contrib-4.2.0/modules ../opencv-4.2.0

# 5. build 目录内，编译
make -j 16

# 6. 安装
make install

# 删除压缩包
cd .. && sudo rm opencv.zip opencv_contrib.zip

编译时候报错：opencv_contrib/modules/xfeatures2d/src/boostdesc.cpp:673:20: fatal error: boostdesc_bgm.i: No such file or directory
opencv_contrib-4.2.0/modules/xfeatures2d/src/boostdesc.cpp:654:20: fatal error: boostdesc_bgm.i: 没有那个文件或目录
解决：网络问题，添加代理，然后重复 4，5 步
参考：https://www.cnblogs.com/arxive/p/11778731.html

Using OpenCV with gcc and CMake

参考：https://docs.opencv.org/4.x/db/df5/tutorial_linux_gcc_cmake.html

// DisplayImage.cpp
#include <stdio.h>
#include <opencv2/opencv.hpp>

using namespace cv;
int main(int argc, char** argv )
{
	if ( argc != 2 )
	{
		printf("usage: DisplayImage.out <Image_Path>\n");
		return -1;
	}
	Mat image;
	image = imread( argv[1], IMREAD_COLOR );
	if ( !image.data )
	{
		printf("No image data \n");
		return -1;
	}
	namedWindow("Display Image", WINDOW_AUTOSIZE );
	imshow("Display Image", image);
	waitKey(0);
	return 0;
}

CMakeLists.txt:

cmake_minimum_required(VERSION 2.8)
set(OpenCV_DIR /home/h/.local/lib/cmake/opencv4)
project( DisplayImage )
find_package( OpenCV REQUIRED )
include_directories( ${OpenCV_INCLUDE_DIRS} )
add_executable( DisplayImage DisplayImage.cpp )
target_link_libraries( DisplayImage ${OpenCV_LIBS} )

Create build directory and switch into it

mkdir build && cd build

.
├── build
├── CMakeLists.txt
└── DisplayImage.cpp

cmake ..

报错： By not providing "FindOpenCV.cmake" in CMAKE_MODULE_PATH this project has
asked CMake to find a package configuration file provided by "OpenCV", but
CMake did not find one.
Could not find a package configuration file provided by "OpenCV" with any
of the following names:
OpenCVConfig.cmake
opencv-config.cmake
Add the installation prefix of "OpenCV" to CMAKE_PREFIX_PATH or set
"OpenCV_DIR" to a directory containing one of the above files. If "OpenCV"
provides a separate development package or SDK, be sure it has been
installed.

解决：

方式1：

# CMakeLists.txt:
set(OpenCV_DIR /home/h/.local/lib/cmake/opencv4)

方式2：

cmake -DOpenCV_DIR=/home/h/.local/lib/cmake/opencv4  ..

然后：

# 在build 文件夹下
make
./DisplayImage lena.jpg

报错：

Gtk-Message: 23:48:54.655: Failed to load module "canberra-gtk-module"

解决：

sudo apt-get install libcanberra-gtk-module

其他：

为python3构建opencv：https://stackoverflow.com/a/39409570

参考：https://cloud.tencent.com/developer/article/1657529

安装 opencv with cuda

参考：https://zhuanlan.zhihu.com/p/640084627

1. 卸载 opencv

找到之前装opencv的build文件夹，进入后执行卸载指令

sudo make uninstall

然后再把之前留在 usr 文件夹里的剩余文件都删掉。我之前直接按照官网说明安装的，所以东西都在 usr/local/ 下

//需要根据自己的情况修改
sudo rm -r /usr/local/include/opencv4 /usr/local/share/opencv4 /usr/local/lib/libopencv*

注意：我们这里是 /home/h/.local 替换掉 /usr/local/

2. 下载 opencv

https://github.com/opencv/opencv/archive/4.5.4.zip

https://github.com/opencv/opencv_contrib/archive/4.5.4.zip

解压、放到一个文件夹中。

3. 编译支持GPU加速的OpenCV

切记：注意：如果已经在 build 中编译，每次 cmake 前请把 rm CMakeCache.txt 删了再编译就可。

cd opencv-4.5.4
mkdir build && cd build
cmake -D CMAKE_BUILD_TYPE=RELEASE \
  -D CMAKE_INSTALL_PREFIX=/usr/local \
  -D OPENCV_EXTRA_MODULES_PATH=../../opencv_contrib-4.5.4/modules \
  -D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda \
  -D CUDA_ARCH_BIN=8.6 \
  -D WITH_CUDA=ON  \
  -D OPENCV_DNN_CUDA=ON  \
  -D WITH_CUDNN=ON  \
  -D ENABLE_FAST_MATH=ON  \
  -D CUDA_FAST_MATH=ON  \
  -D WITH_CUBLAS=ON  \
  -D OPENCV_GENERATE_PKGCONFIG=1  \
  -D BUILD_EXAMPLES=ON \
  -D BUILD_NEW_PYTHON_SUPPORT=ON \
  -D BUILD_opencv_python3=ON \
  -D HAVE_opencv_python3=ON  \
  -D OPENCV_ENABLE_NONFREE=ON  \
  -D INSTALL_PYTHON_EXAMPLES=ON \
  -D OPENCV_PYTHON3_INSTALL_PATH=/usr/local/lib/python3.8/dist-packages  \
  -D PYTHON_EXECUTABLE=/usr/bin/python3.8 \
  ..

分行写 -D 有空格，写在一行无空格： -DCUDA_ARCH_BIN

网络问题需要梯子
查询 DCUDA_ARCH_BIN=8.6 ，我的 3060是这个。https://developer.nvidia.com/cuda-gpus#compute

lscpu 查看有多少个内核可用，然后使用。

架构： x86_64
CPU 运行模式： 32-bit, 64-bit
字节序： Little Endian
Address sizes: 48 bits physical, 48 bits virtual
CPU: 12

make -j 12

参考： https://blog.csdn.net/woshicver/article/details/124763031#t4

请注意该教程，不要做软链接

4. 安装 OpenCV

sudo make install -j 12

使用 python3 解释器和 cv2.cuda.printCudaDeviceInfo(0) 来验证库是否正常工作：

5. 清理

安装后，有些文件我们将不再需要，可以安全删除：

cd ..
rm -rf opencv-4.5.4 opencv_contrib-4.5.4

使用的 CMake 标志参考

CMAKE_BUILD_TYPE =RELEASE：禁用调试选项

https://docs.opencv.org/4.x/db/d05/tutorial_config_reference.html#tutorial_config_reference_general_debug

CMAKE_INSTALL_PREFIX =/usr/local: 设置生成的二进制文件的根目录

https://docs.opencv.org/4.x/db/d05/tutorial_config_reference.html#tutorial_config_reference_install_root

WITH_CUDA =ON：启用 CUDA 支持

https://docs.opencv.org/4.x/db/d05/tutorial_config_reference.html#tutorial_config_reference_func_hetero

WITH_CUDNN=ON：（未记录）启用 cuDNN 支持

WITH_CUBLAS=ON：（未记录）启用 cuBLAS 支持

WITH_TBB =ON：使用线程构建块启用并行编程

https://docs.opencv.org/4.x/db/d05/tutorial_config_reference.html#tutorial_config_reference_func_core

OPENCV_DNN_CUDA =ON：启用 CUDA 后端并构建dnn模块（需要 CUDA、cuDNN 和 cuBLAS）

https://docs.opencv.org/4.x/db/d05/tutorial_config_reference.html#tutorial_config_reference_dnn

OPENCV_ENABLE_NONFREE =ON：启用受专利保护的算法

https://docs.opencv.org/4.x/db/d05/tutorial_config_reference.html#tutorial_config_reference_misc

CUDA_ARCH_BIN =8.7：指定NVIDIA GPU架构版本，根据自己的来

https://github.com/opencv/opencv/blob/4.x/cmake/OpenCVDetectCUDA.cmake#L217

OPENCV_EXTRA_MODULES_PATH =$HOME/opencv_contrib/modules：包含 OpenCV 的额外模块

https://docs.opencv.org/4.x/db/d05/tutorial_config_reference.html#tutorial_config_reference_general_contrib

BUILD_EXAMPLES =OFF：禁用构建示例

https://docs.opencv.org/4.x/db/d05/tutorial_config_reference.html#tutorial_config_reference_general_tests

HAVE_opencv_python3 =ON：强制依赖opencv_python3

https://docs.opencv.org/4.x/db/d05/tutorial_config_reference.html#tutorial_config_reference_func

#include <opencv2/opencv.hpp>
#include <vector>
#include <string>
#include <eigen3/Eigen/Core>
#include <unistd.h>
#include <opencv2/cudastereo.hpp>
#include <chrono>
#include <iostream>

using namespace std;

void calcu_disparity(cv::Mat& left, cv::Mat& right, cv::Mat& disparity)
{
	auto t_a = chrono::high_resolution_clock::now();
	cv::Ptr<cv::StereoSGBM> sgbm = cv::StereoSGBM::create(
		0, 96, 9, 8 * 9 * 9, 32 * 9 * 9, 1, 63, 10, 100, 32);
	cv::Mat disparity_sgbm;
	sgbm->compute(left, right, disparity_sgbm);
	disparity_sgbm.convertTo(disparity, CV_32F, 1.0 / 16.0f);
	auto t_b = chrono::high_resolution_clock::now();
	cout<<"no cuda:"<< chrono::duration_cast<chrono::microseconds>(t_b-t_a).count() / 1000<<"ms"<<endl;
}

void calcu_disparity_cuda(cv::Mat& left, cv::Mat& right, cv::Mat& disparity)
{
	auto t_a = chrono::high_resolution_clock::now();
	cv::Ptr<cv::cuda::StereoSGM> sgbm = cv::cuda::createStereoSGM(
		0, 128, 8, 180, 3, cv::cuda::StereoSGM::MODE_HH);
	cv::Mat disparity_sgbm(left.size(), CV_16S);
	cv::cuda::GpuMat cudaDisparityMap(left.size(), CV_16S);
	cv::cuda::GpuMat cudaDrawColorDisparity(left.size(), CV_8UC4);
	cv::cuda::GpuMat cudaLeftFrame, cudaRightFrame;
	cudaLeftFrame.upload(left);
	cudaRightFrame.upload(right);
	sgbm->compute(cudaLeftFrame, cudaRightFrame, cudaDisparityMap);
	cudaDisparityMap.download(disparity_sgbm);
	disparity_sgbm.convertTo(disparity, CV_32F, 1.0 / 16.0f);
	auto t_b = chrono::high_resolution_clock::now();
	cout<<"with cuda:"<< chrono::duration_cast<chrono::microseconds>(t_b-t_a).count() / 1000<<"ms"<<endl;
}

void get_data(int i, std::string& left_path, std::string& right_path)
{
	std::string data_path = "/home/h/projects/C++/2011_09_26_drive_0048_sync/2011_09_26/2011_09_26_drive_0048_sync/";
	std::string left_data_path = data_path+"image_00/data/";
	std::string right_data_path = data_path+"image_01/data/";
	char ss[22];
	sprintf(ss,"%010d",i);
	left_path = left_data_path + ss+".png";
	right_path = right_data_path + ss+".png";
}

int main(int argc, char **argv)
{
	int data_size = 22;
	for(int i = 0;i<data_size;++i){
		std::string left_image_path, right_image_path;
		get_data(i, left_image_path, right_image_path);
		cv::Mat left_image = cv::imread(left_image_path, 0);
		cv::Mat right_image = cv::imread(right_image_path, 0);

		// if (!left_image.data || !right_image.data) {
		//     // 图像加载失败，处理错误逻辑
		//     std::cerr << "Failed to load images." << std::endl;
		// } else {
		//     // 图像加载成功，显示图像
		//     cv::imshow("Left Image", left_image);
		//     cv::imshow("Right Image", right_image);
		//     // cv::waitKey(0); // 等待按键
		// }
		cv::Mat disparity;
		calcu_disparity(left_image, right_image, disparity);
		cv::imshow("disparity Image", disparity);
		cv::Mat disparity_cuda;
		calcu_disparity_cuda(left_image, right_image, disparity_cuda);
		cv::imshow("disparity_cuda Image", disparity_cuda);
		const char c = cv::waitKey(1);
		if (c == 27) // ESC
			break;
	}
	return 0;
}

cmakelist

cmake_minimum_required(VERSION 2.8)
set(OpenCV_DIR /usr/local/lib/cmake/opencv4)
project( main )
find_package( OpenCV REQUIRED )
include_directories( ${OpenCV_INCLUDE_DIRS} )
add_executable( main main.cpp )
target_link_libraries( main ${OpenCV_LIBS} )