交叉编译OpenCV与OpenCV_contrib到ARM的经验分享

继上次完成NumCpp的移植之后，这些我们来进行OpenCV与OpenCV_contrib的移植。

1. OpenCV、OpenCV_contrib源码下载

GitHub - opencv/opencv: Open Source Computer Vision LibraryOpen Source Computer Vision Library. Contribute to opencv/opencv development by creating an account on GitHub.https://github.com/opencv/opencvhttps://github.com/opencv/opencv_contribhttps://github.com/opencv/opencv_contrib

2. 第三方库的准备

交叉编译OpenCV之前，一般需要libz, libjpeg, libpng, yasm, libx264, libxvid, ffmpeg这些第三方库的编译。

具体的编译过程，我就不多说了，基本上就是检查编译器是否为交叉编译器，然后进行编译即可。这里重点注意ffmpeg，一会儿会提到。

3. 编译opencv与opencv_contrib

编译之前，设置后pkg的路径，输入一下命令

export PKG_CONFIG_PATH=/home/xinzhe/Documents/opencv_depend/lib/pkgconfig:$PKG_CONFIG_PATH

首先从GitHub中clone得到两个文件夹之后，进入到opencv文件夹，进行以下操作

mkdir build
cd build
cmake-gui

之后，弹出cmake-gui的界面。

我们首先设置好source code与build的路径，然后点击configure，然后选择Unix Makefiles里的specify options for cross-compiling，然后进入到新的窗口，填入linux、arm、编译器、依赖库路径，点击finish。

然后configure开始运行。

运行结束后，我们从下面的log里，检查Video I/O：这一项，发现里面的ffmpeg处是NO，也就是没有检测到ffmpeg。这个时候我们需要增加一个值，添加OPENCV_ENABLE_PKG_CONFIG变量，并设其值为true，然后我们再configure，此时发现ffmpeg处为YES。

这样，我们编译出来的opencv能够读视频文件，否则不能够读取成功。

然后我们搜索设置以下内容，这里的selected文件夹中的modules是我从opencv_contrib中的modules下截取的，我只使用了opencv_contrib/modules中的bgsem，因此我把它放到selected/modules里。不把整个opencv_contrib/modules设置为extra_modules是因为，其中的某些库还需要另外的第三方库，我就懒得弄了，如果大家想编译整个opencv_contrib/modules，也是可以的。

OPENCV_EXTRA_MODULES_PATH=/home/xinzhe/Documents/selected/modules

并且设置好CMAKE_INSTALL_PREFIX安装路径之后 ，将以下这些值取消勾选

WITH_TIFF
WITH_GTK
WITH_CUDA
WITH_OPENCL
WITH_OPENEXR
WITH_1394

设置CMAKE_BUILD_TYPE=Release

之后，我们重新点击configure，完成后点击Generate生成Makefile。

关闭cmake-gui，并在build目录下

make
make install

这样我们就得到了opencv与opencv_contrib的交叉编译文件。

4. 交叉编译自己的cpp文件

在自己的cpp文件下，创建Compile.cmake与CMakeLists.txt文件。

其中CMakeLists.txt如下

cmake_minimum_required(VERSION 2.8)

Project(CVWorld)

set(CMAKE_FIND_ROOT_PATH /home/xinzhe/Documents/opencv_v853)
find_package(OpenCV REQUIRED)

##下面是输出信息
message(STATUS "Opnecv library status: ")
message(STATUS "> version: ${OpenCV_VERSION} ")
message(STATUS ">libraries: ${OpenCV_LIBS} ")
message(STATUS "> include: ${OpenCV_INCLUDE_DIRS}  ")


include_directories(${OpenCV_INCLUDE_DIRS}) 


add_executable(VideoHitMan main.cpp)

target_link_libraries(VideoHitMan ${OpenCV_LIBS})

然后，我们创建build文件夹，按以往的操作编译，运行可执行文件，发现缺少一些.so动态库文件。

这个时候，我们将opencv_depend/lib下的所有库文件cp到opencv_v853/lib下，这样就可以编译成功了。

5. 在V853开发板上运行编译得到的程序

得到可执行文件后，将其与opencv_v853 push到板子上，然后输入

export LD_LIBRARY_PATH=/root/opencv_v853/lib:$LD_LIBRARY_PATH

运行程序，发现一大堆库有问题。这个时候就重新建立指向链接就行。

最后就可以运行成功啦。

root@TinaLinux:/# cd root
root@TinaLinux:~# export LD_LIBRARY_PATH=/root/opencv_v853/lib:$LD_LIBRARY_PATH
root@TinaLinux:~# ./VideoHitMan
the video is /root/4.mp4
Height:720 Width:1280
This is the start row :226
This is the end row :712
col_mid = 108
Height:720 Width:1280
This is the start row :226
This is the end row :704
col_mid = 108
Height:720 Width:1280
This is the start row :228
This is the end row :687
col_mid = 112
Height:720 Width:1280
This is the start row :229
This is the end row :672
col_mid = 115
Height:720 Width:1280
This is the start row :230
This is the end row :664
col_mid = 116
Height:720 Width:1280
This is the start row :231
This is the end row :663
col_mid = 116
Height:720 Width:1280
This is the start row :235
This is the end row :663
col_mid = 117
Height:720 Width:1280
This is the start row :238
This is the end row :663
col_mid = 117
Height:720 Width:1280
This is the start row :243
This is the end row :662
col_mid = 118
Height:720 Width:1280
This is the start row :246
This is the end row :661
col_mid = 118
Height:720 Width:1280
This is the start row :250
This is the end row :660
col_mid = 120
Height:720 Width:1280
This is the start row :252
This is the end row :656
col_mid = 121
Height:720 Width:1280
This is the start row :255
This is the end row :650
col_mid = 123
Height:720 Width:1280
This is the start row :256
This is the end row :642
col_mid = 126
Height:720 Width:1280
This is the start row :257
This is the end row :634
col_mid = 128
Height:720 Width:1280
This is the start row :257
This is the end row :621
col_mid = 132
Height:720 Width:1280
This is the start row :256
This is the end row :602
col_mid = 139
Height:720 Width:1280
This is the start row :256
This is the end row :601
col_mid = 138
Height:720 Width:1280
This is the start row :257
This is the end row :599
col_mid = 139
Height:720 Width:1280
This is the start row :258
This is the end row :596
col_mid = 140
Height:720 Width:1280
This is the start row :261
This is the end row :595
col_mid = 142
Height:720 Width:1280
This is the start row :263
This is the end row :595
col_mid = 143
Height:720 Width:1280
This is the start row :266
This is the end row :594
col_mid = 143
Height:720 Width:1280
This is the start row :269
This is the end row :594
col_mid = 145
Height:720 Width:1280
This is the start row :271
This is the end row :594
col_mid = 145
Height:720 Width:1280
This is the start row :273
This is the end row :593
col_mid = 146
location = 73216
Generated binary tensor from silhouette
root@TinaLinux:~# 

今天就到这里啦，感谢看到的每个小伙伴！