交叉编译OpenCV2.3.1

不念不忘少年蓝@ 2022-01-14 20:51 497阅读 0赞

[2019独角兽企业重金招聘Python工程师标准>>> ][2019_Python_] ![hot3.png][]

编译这个库真他麻烦，明明工作重点不是编译OpenCV库，但是时间就偏偏耗在这里，从昨天晚上到现在终于编译成功，说说编译的过程。

我使用的环境linux mint 不记得几了。

编译器arm-linux-gnueabihf-gcc/g++ 4.6 (支持硬件浮点)，

目标板是 cubieboard开发板 1ＧHz的ＣＰＵ（支持硬件浮点）,1GRAM，目前算是很强悍一款开发板了,而且价格公道358元,可以在上面运行opencv，杠杆的。买它回来就是在为了在上面做视觉处理的。

我的编译过程是参考csk前辈的编译过程[http://www.csksoft.net/blog/post/mk802\_dev\_opencv.htm][http_www.csksoft.net_blog_post_mk802_dev_opencv.htm]l，但是他的过程有些小问题，要真正试过才知道![13.gif][]。

这里我放上我编译的opencv库，以及下面要用到的png zlib jpeg依赖库

[ http://vdisk.weibo.com/s/vMeNN/1364722558][http_vdisk.weibo.com_s_vMeNN_1364722558]

Hi，推荐文件给你 "zlib\_jpeg\_png.tar.gz"[ http://vdisk.weibo.com/s/vMeQd/1364722532][http_vdisk.weibo.com_s_vMeNN_1364722558]

1.下载硬件浮点编译器，

sudo apt-get install gcc-arm-linux-gnueabihf g++-arm-linux-gnueabihf

2.编译依赖库

2.1 先设定几个环境变量方便后面的编译

export COMPILE\_PREFIX=arm-linux-gnueabihf-  
export INSTALL\_PREFIX=\`pwd\`/crosscompile\_root  
export DL\_FOLDER=\`pwd\`/dl  
export BUILD\_FOLDER=\`pwd\`/buildpool  
export CC=$\{COMPILE\_PREFIX\}gcc  
  
同时建立上面的几个目录用于存放编译过程的文件,把下载到的zlib/jpeg/png库放到dl目录中。

mkdir crosscompile\_root dl buildpool

2.2 编译zlib，可以把下面的命令制作成一个脚本文件

pushd $BUILD\_FOLDER  
tar xf $DL\_FOLDER/zlib-1.2.7.tar.gz  
pushd zlib-1.2.7  
./configure --prefix=$INSTALL\_PREFIX  
make  
make install  
popd

2.3 编译jpeg

tar xf $DL\_FOLDER/jpegsrc.v8d.tar.gz  
pushd jpeg-8d  
./configure --host=arm-linux-gnueabihf --prefix=$INSTALL\_PREFIX  
mkdir -p $INSTALL\_PREFIX/bin  
mkdir -p $INSTALL\_PREFIX/man/man1  
make  
make install  
popd  
  
2.4 编译png

tar xf $DL\_FOLDER/libpng-1.5.4.tar.gz  
pushd libpng-1.5.4  
./configure --host=arm-linux-gnueabihf --prefix=$INSTALL\_PREFIX CPPFLAGS=-I$INSTALL\_PREFIX/include LDFLAGS=-L$INSTALL\_PREFIX/lib  
make  
make install  
popd

2.5 配置编译opencv

完成上面的几步应该已经在crosscompile\_root/lib目录下有相应的库了

cd ~/OpenCV2.3.1/

mkdir build

这build目录里建立toolchain.cmake文件，输入以下内容

set( CMAKE\_SYSTEM\_NAME Linux )  
set( CMAKE\_SYSTEM\_PROCESSOR arm )  
set( CMAKE\_C\_COMPILER arm-linux-gnueabihf-gcc )  
set( CMAKE\_CXX\_COMPILER arm-linux-gnueabihf-g++ )  
set( CMAKE\_LIBRARY\_PATH ~/crosscompile\_root/lib $\{CMAKE\_LIBRARY\_PATH\})  
set( CMAKE\_INCLUDE\_PATH ~/crosscompile\_root/include $\{CMAKE\_INCLUDE\_PATH\})

然后执行命令，csk前辈在这里有个小错误，我已经改了

cmake -DCMAKE\_INSTALL\_PREFIX=$INSTALL\_PREFIX -DZLIB\_INCLUDE\_DIR=$INSTALL\_PREFIX/lib -DPNG\_INCLUDE\_DIR=$INSTALL\_PREFIX/lib -DJPEG\_INCLUDE\_DIR=$INSTALL\_PREFIX/lib -DOPENCV\_EXTRA\_C\_FLAGS="-mfpu=neon -I$INSTALL\_PREFIX/include -L$INSTALL\_PREFIX/lib -Wl, -rpath,$INSTALL\_PREFIX/lib" -DCMAKE\_TOOLCHAIN\_FILE=toolchain.cmake ..

以下是我的cmake结果，可以看到已经禁掉好多东西，因为是嵌入式开发可以吧一些不需要用到的模块去掉，

Extracting svn version, please wait...  
\-- SVNVERSION: svn:已导出  
\-- Detected version of GNU GCC: 46 (406)  
\-- Could NOT find TIFF (missing: TIFF\_LIBRARY)  
\-- Could NOT find Jasper (missing: JASPER\_LIBRARY JASPER\_INCLUDE\_DIR)  
\-- checking for module 'libv4l1'  
\-- package 'libv4l1' not found  
\-- Use NumPy headers from: /usr/lib/python2.7/dist-packages/numpy/core/include  
Traceback (most recent call last):  
File "<string>", line 1, in <module>  
ImportError: No module named sphinx  
\-- Parsing 'cvconfig.h.cmake'  
\--  
\-- General configuration for opencv 2.3.1 =====================================  
\--  
\-- Built as dynamic libs?: YES  
\-- Compiler: /usr/bin/arm-linux-gnueabihf-g++  
\-- C++ flags (Release): -Wall -pthread -ffunction-sections -mfpu=neon -I/home/msg/crosscompile\_root/include -L/home/msg/crosscompile\_root/lib -Wl, /home/msg/crosscompile\_root/lib -O3 -DNDEBUG -fomit-frame-pointer -DNDEBUG  
\-- C++ flags (Debug): -Wall -pthread -ffunction-sections -mfpu=neon -I/home/msg/crosscompile\_root/include -L/home/msg/crosscompile\_root/lib -Wl, /home/msg/crosscompile\_root/lib -g -O0 -DDEBUG -D\_DEBUG -ggdb3  
\-- Linker flags (Release):  
\-- Linker flags (Debug):  
\--  
\-- GUI:  
\-- GTK+ 2.x: NO  
\-- GThread: NO  
\--  
\-- Media I/O:  
\-- ZLib: YES  
\-- JPEG: TRUE  
\-- PNG: TRUE  
\-- TIFF: build  
\-- JPEG 2000: build  
\-- OpenEXR: NO  
\-- OpenNI: NO  
\-- OpenNI PrimeSensor Modules: NO  
\-- XIMEA: NO  
\--  
\-- Video I/O:  
\-- DC1394 1.x: NO  
\-- DC1394 2.x: NO  
\-- FFMPEG: NO  
\-- codec: NO  
\-- format: NO  
\-- util: NO  
\-- swscale: NO  
\-- gentoo-style: YES  
\-- GStreamer: NO  
\-- UniCap: NO  
\-- PvAPI: NO  
\-- V4L/V4L2: /1  
\-- Xine: NO  
\--  
\-- Other third-party libraries:  
\-- Use IPP: NO  
\-- Use TBB: NO  
\-- Use ThreadingFramework: NO  
\-- Use Cuda: NO  
\-- Use Eigen: YES  
\--  
\-- Interfaces:  
\-- Python: NO  
\-- Python interpreter: /usr/bin/python -B (ver 2.7)  
\-- Python numpy: YES  
\-- Java: NO  
\--  
\-- Documentation:  
\-- Sphinx: NO  
\-- PdfLaTeX compiler: NO  
\-- Build Documentation: NO  
\--  
\-- Tests and samples:  
\-- Tests: YES  
\-- Examples: NO  
\--  
\-- Install path: /home/msg/crosscompile\_root  
\--  
\-- cvconfig.h is in: /home/msg/OpenCV-2.3.1/build  
\-- -----------------------------------------------------------------  
\--  
\-- Configuring done  
\-- Generating done  
\-- Build files have been written to: /home/msg/OpenCV-2.3.1/build

///

然后使用ccmake 或者cmake-gui微调，

ccmake .

![175248_sSD6_998193.png][]

加入了-mfpu=neon 据说是可以在cubieboard的ＣＰＵ上进行优化，在ccmake中去掉python,gtk,cuba,1394lib等，因为在arm-linux-gcc编译器里面没有这些相应的库，我就是裁在这里.

3 .直接安装opencv

其实cubieboard上的linaro Ubuntu可以直接从apt-get 安装arm版的opencv,

sudo apt-get install libopencv-\*

3.1 测试opencv

#include <opencv2/opencv.hpp>
    #includee <stdio.h>
     int main( int argc, char * argv[] ) {
        // grab one frame from the camera specified via arg[1]
        int camera_id = 0;
    
        if (argc > 1) camera_id = atoi(argv[1]);
    
        cv::VideoCapture cap(camera_id);
    
        if(!cap.isOpened())  // check if we succeeded
        {
            printf("failed to open the camera with id %d.\n", camera_id);
            return -1;
        }
       
        // capture one frame
        cv::Mat frame;
        cap >> frame; // get a new frame from camera
       
        // save to file
        imwrite("camera_captured.png", frame);
    
        return 0;
    }

编译以上程序

arm-linux-gnueabihf-g++ main.cpp -o main -L/home/msg/crosscompile\_root/lib -I/home/msg/crosscompile\_root/include -lpthread -lopencv\_core -lopencv\_imgproc -lopencv\_highgui -lopencv\_ml -lopencv\_video -lopencv\_features2d -lopencv\_calib3d -lopencv\_objdetect -lopencv\_contrib -lopencv\_legacy -lopencv\_flann -rpath-link  
  
3.2 把opencv库放到开发板上

复制到ＳＤ卡，由于我的是校园网用锐捷的设备，因此搞不到无线网络，也不能用cubieboard上网，很悲剧

sudo cp -r ~/crosscompile/lib/\* /media/yoursdcard/usr/local/lib

sudo cp -r ~/main /media/yoursdcard/home/linaro

3.3 执行

./main 0 //0为摄像头编号

得到 png图像

camera_captured.png

转载于:https://my.oschina.net/elinuxer/blog/118877

[2019_Python_]: https://my.oschina.net/u/2663968/blog/3061697
[hot3.png]: /images/20220114/a5a6999eebc849e092c2794c417b74e9.png
[http_www.csksoft.net_blog_post_mk802_dev_opencv.htm]: http://www.csksoft.net/blog/post/mk802_dev_opencv.htm
[13.gif]: /images/20220114/bbdfbf25c0154820865e46270503a89a.png
[http_vdisk.weibo.com_s_vMeNN_1364722558]: http://vdisk.weibo.com/s/vMeNN/1364722558
[175248_sSD6_998193.png]: /images/20220114/ea28e024ac504402bd442982bc298533.png