Ubuntu 14.04: install OpenCV with CUDA

CUDA Toolkit

下载：https://developer.nvidia.com/cuda-downloads

安装：

• 方法一：使用distribution-specific packages 安装 CUDA，ubuntu对应deb安装包
• 方法二：使用distribution-independent package 安装 CUDA，即run安装包

推荐使用方法一安装，比较简单快键，只要以下几条命令即可。方法二见参考。
1.Install repository meta-data
sudo apt-get install build-essential
sudo dpkg -i cuda-repo-<distro>_<version>_<architecture>.deb
2.Update the Apt repository cache
sudo apt-get update
3.Install CUDA
sudo apt-get install cuda
4.Environment Setup
修改etc/profile或~/.bashrc，source /etc/profile（对应profile）或sudo ldconfig（对应bashrc）使之生效
export PATH=/usr/local/cuda-7.0/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-7.0/lib64:$LD_LIBRARY_PATH
5.Install Writable Samples
cuda-install-samples-7.0.sh <dir>
6.Reboot
重启系统

参考：
- Getting Started Linux :: CUDA Toolkit Documentation
- Ubuntu12.04配置NVIDIA cuda5.5经验帖 - Rachel Zhang的专栏 - 博客频道 - CSDN.NET
- Caffe + Ubuntu 15.04 + CUDA 7.0 新手安装配置指南
- fbcunn/INSTALL.md at master · facebook/fbcunn · GitHub

OpenCV

1.Install required packages

sudo apt-get update

sudo apt-get install build-essential checkinstall cmake pkg-config yasm libtiff4-dev libjpeg-dev libjasper-dev libavcodec-dev libavformat-dev libswscale-dev libdc1394-22-dev libxine-dev libgstreamer0.10-dev libgstreamer-plugins-base0.10-dev libv4l-dev python-dev python-numpy libtbb-dev libqt4-dev libgtk2.0-dev libfaac-dev libmp3lame-dev libopencore-amrnb-dev libopencore-amrwb-dev libtheora-dev libvorbis-dev libxvidcore-dev x264 v4l-utils libeigen3-dev 

sudo add-apt-repository ppa:jon-severinsson/ffmpeg  
sudo apt-get update  
sudo apt-get install ffmpeg  
sudo apt-get install frei0r-plugins  

2.Build and install OpenCV

mkdir release  
cd release  
cmake -D WITH_TBB=ON -D BUILD_NEW_PYTHON_SUPPORT=ON -D WITH_V4L=ON -D INSTALL_C_EXAMPLES=OFF -D INSTALL_PYTHON_EXAMPLES=OFF -D BUILD_EXAMPLES=OFF -D WITH_QT=ON -D WITH_OPENGL=ON -D ENABLE_FAST_MATH=1 -D WITH_CUDA=ON -D CUDA_FAST_MATH=1 -D WITH_CUBLAS=1 -D CUDA_GENERATION=Auto -D WITH_GSTREAMER_0_10=OFF ..  

检查cmake的输出，确认CUDA和CUBLAS将被安装

--     Use Cuda:                    YES (ver 6.5)
--     Use OpenCL:                  YES
-- 
--   NVIDIA CUDA
--     Use CUFFT:                   YES
--     Use CUBLAS:                  YES
--     USE NVCUVID:                 NO
--     NVIDIA GPU arch:             20
--     NVIDIA PTX archs:            
--     Use fast math:               YES

If everything is correct you can install OpenCV:

make  
sudo make install  

如果使用pkgconfig，需要修改环境变量：

echo '/usr/local/lib' | sudo tee -a /etc/ld.so.conf.d/opencv.conf  
sudo ldconfig  
printf '# OpenCV\nPKG_CONFIG_PATH=$PKG_CONFIG_PATH:/usr/local/lib/pkgconfig\nexport PKG_CONFIG_PATH\n' >> ~/.bashrc  
source ~/.bashrc  

注意：

• 其中有几个example没有编译，改为OFF
• -D WITH_GSTREAMER_0_10=OFF是为了防止出现error: ‘GstMiniObjectClass’ does not name a type
• -D CUDA_GENERATION=Auto是为了防止出现nvcc fatal : Unsupported gpu architecture 'compute_11'

参考：
- Ubuntu 14.04: install OpenCV with CUDA
- Build OpenCV with CUDA support - Stack Overflow
- Ubuntu 编译opencv出错：‘GstMiniObjectClass’ does not name a type 解决方法 - jdpshq的专栏 - 博客频道 - CSDN.NET

Test

测试一下OpenCV的功能，如opencv-2.4.10/samples/gpu/中的surf_keypoint_matcher.cpp
这里提供两种方法：

• 使用cmake建立工程
  新建CMakeLists.txt，内容如下：

PROJECT(test)
CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
FIND_PACKAGE( OpenCV REQUIRED )
ADD_EXECUTABLE(surf surf_keypoint_matcher.cpp)
TARGET_LINK_LIBRARIES(surf ${OpenCV_LIBS}) 

mkdir build
cd build
cmake ..
./surf --left ../tsucuba_left.png --right ../tsucuba_right.png 

• 使用命令行g++编译

g++ surf_keypoint_matcher.cpp -o surf `pkg-config --cflags --libs opencv` -L/usr/local/cuda/lib64
./surf --left tsucuba_left.png --right tsucuba_right.png 

注：
不加上-L/usr/local/cuda/lib64会出错，因为在/usr/local/lib/pkgconfig/opencv.pc中并没有CUDA library的信息。

/usr/bin/ld: cannot find -lcufft
/usr/bin/ld: cannot find -lcublas
/usr/bin/ld: cannot find -lnpps
/usr/bin/ld: cannot find -lnppi
/usr/bin/ld: cannot find -lnppc
/usr/bin/ld: cannot find -lcudart