openCV安装

2019-10-21

AS和cmake笔记

android的cpu架构

• android查看cpu架构(termux)

cat /proc/cpuinfo

• AArch32是ARMv7架构的一种执行状态
• AArch64是ARMv8架构的一种执行状态
  • 参考
  • 参考
• 64位ARMv8包含上面两种状态
  • 参考
  • 知乎
  • 知乎

cmake在AS中的使用

不要把mk和cmake两种方法混淆

输出控制(message)

(TODO)如果想直接看到打印信息,使用WARNING以上的级别进行打印

• message(WARNING "fuck")
  • 在Build->Build Output->CMake warnings查看
  • 在Build->Build Output->Toggle View查看

Android

Android Studio版本为3.5

只使用java模块

3.x-3.4.7-3.x版本

1. File->New->Import Module->选择opencv文件夹下的sdk/java文件夹,此时Android Studio会自动加载Module name
2. File->Project Structure->Dependencies->选择app->添加Module Dependency->选择opencv
3. 将opencv文件夹下的sdk/native/libs下自己手机cpu对应的架构文件夹复制到app/libs下
4. 将opencv模块的build.gradle的complieSdkVersion,buildToolVersion,minSdkVersion,targetSdkVersion改成和app的build.gradle一样
5. 在app的build.gradle的android节点下添加

   sourceSets { main { jni.srcDirs = [] jniLibs.srcDirs = ['libs'] } }

然后点击sync now

6. 在使用opencv前进行初始化

   static { if (!OpenCVLoader.initDebug()) { ... } }

常见错误

1. Caused by: CvException [org.opencv.core.CvException: OpenCV(4.1.1) /build/master_pack-android/opencv/modules/java/generator/src/cpp/utils.cpp:101: error: (-215:Assertion failed)

在Mat与Bitmap互转的时候长宽设置不对

只使用native模块

3.x-3.4.7-3.x版本

方法一

1. CMakeLists.txt编写
   1. include头文件目录
   2. 加载libopencv动态库(libopencv_world或libopencv_java3),命名为opencv_lib
   3. 将opencv_lib加载至目标文件

set(opencv_dir /home/lynx/fuck_mount/opencv/build_dir/4_1_0_shared/OpenCV-android-sdk)
set(opencv_lib libopencv_world.so) # 或libopencv_java3.so

include_directories(${opencv_dir}/sdk/native/jni/include)
add_library(opencv_lib SHARED IMPORTED)
set_target_properties(opencv_lib PROPERTIES IMPORTED_LOCATION
    ${opencv_dir}/sdk/native/libs/${ANDROID_ABI}/${opencv_lib})

target_link_libraries(
        # 目标文件

        opencv_lib
        # 其他库 )

2. 修改app的build.gradle
   • 在android节点下插入,否则会报错找不到libopencv_world.so或libopencv_java3.so

sourceSets {
  main {
    jniLibs.srcDirs = ['/sdk/native/libs 对应目录']
  }
}

方法二

参考:CSDN

1. CMakeLists.txt
   1. 设置OpenCV查找的目录
   2. 查找OpenCV
   3. include头文件目录
   4. 将OpenCV的库添加至目标文件

set(sdk_dir /home/lynx/fuck_mount/opencv/build_dir/4_1_0_shared)
set(OpenCV_DIR ${sdk_dir}/OpenCV-android-sdk/sdk/native/jni)
find_package(OpenCV REQUIRED)
include_directories(${sdk_dir}/OpenCV-android-sdk/sdk/native/jni/include)

target_link_libraries( 
                       # 目标文件

                       ${OpenCV_LIBS}
                       # 其他库 )

• 如果不设置OpenCV_DIR将无法查找OpenCV包

2. 修改app的build.gradle

sourceSets {
  main {
    jniLibs.srcDirs = ['/home/lynx/fuck_mount/opencv/build_dir/4_1_0_shared/OpenCV-android-sdk/sdk/native/libs']
  }
}

• 此外,务必将minSdkVersion调至21以上,否则将会只有v8a架构才能成功编译
• 在app的build.gradle中:minSdkVersion->ANDROID_PLATFORM
• 在ndk的android.toolchain.cmake中:ANDROID_PLATFORM->ANDROID_PLATFORM_LEVEL->ANDROID_NATIVE_API_LEVEL
• 如果不设置minSdkVersion,由于以上的确定关系,ANDROID_NATIVE_API_LEVEL会被设置为16,find_package会失败

4.x-4.1.0-4.x版本

方法一

同3.x版本的方法一

方法二

同3.x版本的方法二

常见问题

1. java.lang.UnsatisfiedLinkError: dlopen failed: library "libc++_shared.so" not found(4.x版本)

stackoverflow

(不确定的解决方法)在app的gradle中修改

externalNativeBuild {
  cmake {
    cppFlags ""
      arguments "-DANDROID_STL=c++_shared"
  }
}

不要使用gnustl_shared或gnustl_static,已经过时

2. java.lang.UnsatisfiedLinkError: dlopen failed: library "libopencv_java3.so" not found(或libopencv_java4.so)

这是由于忘记了上面的第2步.Android Studio默认的jni路径为app/src/main/jniLibs,除了在CMakeLists.txt中导入opencv的库以外还需要二选一

• 将对应版本opencv的库(libs/下一级的各个${ANDROID_ABI}文件夹)移动到jniLibs下
• 在build.gradle中修改默认jni路径(见上面第2步)

3. 各种undefined reference to *(3.x版本)

如std::ios_base::Init::Init()等

TODO

4. 各类AndroidBitmap的问题(3.x版本)

undefined reference to `AndroidBitmap_getInfo'

在CMakeLists.txt增加

target_link_libraries(
        # TODO 解决 AndroidBitmap 报错
        jnigraphics)

参考:CSDN,这是一个很小的库,展示一个稳定的,基于C语言的接口,使本机代码安全地访问Java对象的像素缓冲区的位图

source code(使用完整代码)

3.x-3.x版本(未成功,需要eclipse)

4.x-4.1.0-4.x版本

参考:CSDN

1. 环境变量

   • JAVA_HOME

     export JAVA_HOME=/home/lynx/fuck_mount/Android_Studio/jre # 使用Android Studio的JDK位置

   • ANDROID_SDK(非必须)

     export ANDROID_SDK=/home/lynx/fuck_mount/Android_SDK

   • ANDROID_NDK(非必须)

     export ANDROID_NDK=/home/lynx/fuck_mount/Android_SDK/ndk/{版本}

2. 调整结构

   • 修改toolchain:android ndk的ndk/build/cmake/android.toolchain.cmake(约112行)

     elseif(ANDROID_TOOLCHAIN_NAME MATCHES "-[0-9].[0-9]$") set(ANDROID_TOOLCHAIN clang) # 将gcc改成clang endif()

   • android stl,在源代码的/platforms/android/build_sdk.py(约113行)(非必须)

     self.cmake_vars = dict( ANDROID_STL="c++_static", # change `gnustl_static` to `c++_static` or `c++_shared` ANDROID_ABI=self.name, ANDROID_PLATFORM_ID=platform_id, )

   • 在Android Studio的android sdk文件夹下,创建软连接ln -s build-tools/{版本} tools,否则无法编译安卓模块

3. 修改cmake参数(2选1,推荐world)

   • java
     • OPENCV_ENABLE_NONFREE:改为ON
     • BUILD_SHARED_LIBS:不动或者改为OFF,与生成java库冲突
     • BUILD_ANDROID_PROJECTS:不动或者改为ON
     • BUILD_FAT_JAVA_LIB:不动或者改为ON,保证生成libopencv_java4.so
     • BUILD_JAVA:不动或者改为ON
     • 对/samples/android/CMakeLists.txt中除了ocv_warnings_disable(CMAKE_CXX_FLAGS -Wmissing-declarations)那一行,其他全部注释掉,否则android sdk会在编译sample时报错
   • world
     • OPENCV_ENABLE_NONFREE:改为ON
     • BUILD_SHARED_LIBS:改为ON
     • BUILD_ANDROID_PROJECTS:不动或者改为ON
     • BUILD_FAT_JAVA_LIB:改为OFF,生成libopencv_world.so
     • BUILD_JAVA:不动或者改为ON
     • 对/samples/android/CMakeLists.txt中除了ocv_warnings_disable(CMAKE_CXX_FLAGS -Wmissing-declarations)那一行,其他全部注释掉,否则android sdk会在编译sample时报错

4. 执行platforms/android/build_sdk.py,完成编译

./build_sdk.py --ndk_path {ndk路径} --sdk_path {android sdk路径} --extra_modules_path {path_to_contrib/modules} {目标位置} {opencv源码位置}
# example
/home/lynx/fuck_mount/opencv_4_1_0/opencv-4.1.0/platforms/android/build_sdk.py --ndk_path /home/lynx/fuck_mount/Android/SDK/ndk/18.1.5063045 --sdk_path /home/lynx/fuck_mount/Android/SDK --extra_modules_path /home/lynx/fuck_mount/opencv/source_code/opencv_contrib_4_1_0/modules /home/lynx/fuck_mount/opencv_4_1_0/android_build /home/lynx/fuck_mount/opencv_4_1_0/opencv-4.1.0

如果没有设置则默认为环境变量

• sdk-path:Android Studio的SDK目录
• ndk-path:Android Studio的SDK目录下ndk/{ndk版本号文件}

5. 后续关于CMakeLists.txt和build.gradle的修改同native方法

常见错误

1. java.lang.UnsatisfiedLinkError: dalvik.system.PathClassLoader[DexPathList[[zip file "/data/app/com.example.stitch-sPWgBONIqUrdX9XLB1OY6w==/base.apk"],nativeLibraryDirectories=[/data/app/com.example.stitch-sPWgBONIqUrdX9XLB1OY6w==/lib/arm64, /data/app/com.example.stitch-sPWgBONIqUrdX9XLB1OY6w==/base.apk!/lib/arm64-v8a, /system/lib64, /product/lib64]]] couldn't find "libnative-lib.so"

缺少相应ABI的so库(目前计时在gradle里加入abiFilters也没有用),上述报错是因为没有编译全ABI.ABI配置:opencv-4.x/platforms/android/ndk-18-api-level-21.config.py

2. 没装cmake

Executing: ['ninja', '--version']
Failed: [Errno 2] No such file or directory

Android Studio需要在SDK Manager里面预先安装CMake,否则编译的时候会找不到ninja等程序

3. 没有libopencv_java4.so

   • BUILD_FAT_JAVA_LIB和BUILD_ANDROID_PROJECTS要设置ON(或默认)
   • 记得修改android sdk tools文件结构
   • 不能同时编译动态库

4. 编译错误

Regex: 'command line option .* is valid for .* but not for C\+\+'

TODO,可能是编译器的问题或cmake参数设置错误

5. android project报错

A problem occurred configuring project ':face-detection'.
> NDK not configured.
  Download it with SDK manager.

把samples/android/CMakeLists.txt对应代码注释掉

6. ndk报错

CMake Error at /home/lynx/fuck_mount/Android_SDK/ndk/20.1.5948944/build/cmake/android.toolchain.cmake:169 (message):
  GCC is no longer supported.  See
  https://android.googlesource.com/platform/ndk/+/master/docs/ClangMigration.md.

需要将ndk的build/cmake/android.toolchain.cmake中set(ANDROID_TOOLCHAIN gcc)改为clang

7. ninja报错

CMake Error: CMake was unable to find a build program corresponding to "Ninja".  CMAKE_MAKE_PROGRAM is not set.  You probably need to select a different build tool.

将ndk的bin目录(包含ninja等程序)添加至$PATH

8. ccache未安装(3.x版本,TODO:不确定是否为必须)

用apt安装ccache即可

9. ld报错(3.x版本)

cannot open crtbegin_dynamic.o: 没有那个文件或目录
cannot open crtend_android.o: 没有那个文件或目录
cannot find -lstdc++
cannot find -lm
cannot find -lc
cannot find -ldl

TODO修改gcc参数(但很大程度上是由于cmake参数配置错误)

1. 添加-nostdlib
2. 修改--sysroot参数

10. clang++报错

error while loading shared libraries: libtinfo.so.5: cannot open shared object file: No such file or directory

使用apt安装libtinfo5

11. 编译错误

error: use of undeclared identifier 'vcvt_f16_f32'
    float16x4_t v_dst = vcvt_f16_f32(v_src);

换ndk

12. libopencv_java3.so没有生成(3.x版本)

TODO,很大程度上是由于cmake参数配置错误

13. cmake报错

Android SDK: Can't build Android projects as requested by
  BUILD_ANDROID_PROJECTS=ON variable.

• 4.x:android sdk build tool版本过低
• 3.x:需要sdk整合ant(ide为eclipse)

14. java报错

opencv init failed

检查gradle的lib路径是否正确加载

额外说明

4.x版本

• 不需要额外安装的软件:jdk,ninja-build,ndk,直接使用android studio的即可
• 成功编译的版本:
  • ndk:20.1.5948944
  • sdk tools:28.0.3
  • cmake:3.13.4
• 失败的版本:
  • ndk:21.0.6113669(报错undefined referrence to '__write_chk')
  • sdk tools:25.0.3(不能configure android project)

3.x版本 目前还没能成功编译android project

• 3.x成功编译的版本:
  • 无
• 失败的版本:
  • ndk:20.1.5948944(报错undefined referrence to '__vsnprintf_chk')

使用JavaCV

4.x版本

不同导入方法,import的class位置可能不同

opencv部分功能的初始化

github原文

Loader.load(opencv_java.class);// 不能直接放在class开头

使用远程导入

app的build.gradle示例(最外节点)

ext {
    versions = [
            'javacv': '1.4.4',
            'ffmpeg': '4.1',
            'opencv': '4.0.1'
    ]
}

dependencies {
    ...
    implementation(group: 'org.bytedeco', name: 'javacv-platform', version: versions.javacv) {
      exclude group: 'org.bytedeco.javacpp-presets'
    }
    implementation group: 'org.bytedeco.javacpp-presets', name: 'ffmpeg', version: "${versions.ffmpeg}-${versions.javacv}"
    implementation group: 'org.bytedeco.javacpp-presets', name: 'ffmpeg', version: "${versions.ffmpeg}-${versions.javacv}", classifier: 'android-arm'
    implementation group: 'org.bytedeco.javacpp-presets', name: 'ffmpeg', version: "${versions.ffmpeg}-${versions.javacv}", classifier: 'android-arm64'
    implementation group: 'org.bytedeco.javacpp-presets', name: 'opencv', version: "${versions.opencv}-${versions.javacv}"
    implementation group: 'org.bytedeco.javacpp-presets', name: 'opencv', version: "${versions.opencv}-${versions.javacv}", classifier: 'android-arm'
    implementation group: 'org.bytedeco.javacpp-presets', name: 'opencv', version: "${versions.opencv}-${versions.javacv}", classifier: 'android-arm64'
}

作为lib导入

根据手机具体的cpu架构进行导入

app的build.gradle

dependencies {
    implementation fileTree(dir: 'libs', include: ['*.jar'])
}

将文件(视具体情况)

• javacpp.jar
• javacv.jar
• openblas.jar
• openblas-android-arm64.jar
• opencv.jar
• opencv-android-arm64.jar
  复制到app/libs/文件夹里

常见错误

1. cpu架构错误/缺少(需添加-android-arm之类的库)

2019-10-27 16:01:32.439 29256-29322/? E/AndroidRuntime: FATAL EXCEPTION: AsyncTask #1
    Process: com.github.crazyorr.ffmpegrecorder, PID: 29256
    java.lang.RuntimeException: An error occurred while executing doInBackground()

2. 没有对opencv初始化

2019-10-27 16:00:43.930 28933-28933/? E/AndroidRuntime: FATAL EXCEPTION: main
    Process: com.example.javacv, PID: 28933
    java.lang.UnsatisfiedLinkError: No implementation found for long org.opencv.core.Mat.n_Mat() (tried Java_org_opencv_core_Mat_n_1Mat and Java_org_opencv_core_Mat_n_1Mat__)

3. 缺少openblas的相关jar库

2019-10-27 15:59:23.158 28618-28618/? E/AndroidRuntime: FATAL EXCEPTION: main
    Process: com.example.javacv, PID: 28618
    java.lang.NoClassDefFoundError: Failed resolution of: Lorg/bytedeco/openblas/presets/openblas;

混用java+native模块

4.x native(或source code)+3.x java

参考:CSDN

1. 加载native模块

2. File->New->Import Module->选择opencv文件夹下的sdk/java文件夹,自己定义模块的名字

3. 修改模块的gradle文件(TODO)

// 将原来的application改成library
apply plugin: 'com.android.library'

android {
  // 将版本信息改得和`app`的`build.gradle`一样
  compileSdkVersion 29
  buildToolsVersion "29.0.2"
  defaultConfig {
    // 删除applicationId
    minSdkVersion 15
    targetSdkVersion 29
    versionCode 1
    versionName "1.0"
    testInstrumentationRunner "androidx.test.runner.AndroidJUnitRunner"
  }
  // 不改动
  buildTypes {
    release {
      minifyEnabled false
        proguardFiles getDefaultProguardFile('proguard-android.txt'), 'proguard-rules.txt'
    }
  }
}

4. 修改app的gradle文件,加载两者的lib,缺一不可

sourceSets {
  main {
    jniLibs.srcDirs = ['/home/lynx/fuck_mount/opencv/source_code/opencv_3_4_7_android/sdk/native/libs',
              '/home/lynx/fuck_mount/opencv/build_dir/4_1_0_shared/OpenCV-android-sdk/sdk/native/libs']
  }
}

5. 修改app的gradle文件,添加java模块依赖

dependencies {
    implementation project(path: ':openCVLibrary347')
}

Linux

OpenCV source code 3.x(3.4.7)

gui安装

和4.x一样

常见错误

1. fatal error: Eigen/Core: 没有那个文件或目录``#include <Eigen/Core>

   1. 修改WITH_EIGEN=OFF

   2. (或)

      • 先保证安装eigen:sudo apt install libeigen3-dev
      • 添加EIGEN_INCLUDE_PATH=/usr/include/eigen3(视具体情况)

2. ippicv_linux_20141027.tgz下载失败

   CMake Error at 3rdparty/ippicv/downloader.cmake:71 (file): file DOWNLOAD HASH mismatch for file: [/home/lynx/fuck_mount/opencv_3_0_0/opencv/3rdparty/ippicv/downloads/linux-8b449a536a2157bcad08a2b9f266828b/ippicv_linux_20141027.tgz] expected hash: [8b449a536a2157bcad08a2b9f266828b] actual hash: [5c4f36bf7b2421d52289f0297ba1406f] status: [28;"Timeout was reached"]

TODO

3. ippicv_2019_lnx_intel64_general_20180723.tgz下载超时

   IPPICV: Download: ippicv_2019_lnx_intel64_general_20180723.tgz CMake Warning at cmake/OpenCVDownload.cmake:193 (message): IPPICV: Download failed: 28;"Timeout was reached"

   1. 在opencv/opencv_3rdparty的ippicv对应分支中下载对应的tgz

   2. 将tgz放入/home/lynx/Downloads/

   3. 修改opencv的source code: vi /3rdparty/ippicv/ippicv.cmake,将ippicv的https地址改为file:/home/lynx/Downloads/

   4. 重新cmake-gui

OpenCV source code 4.x(4.1.0)

依赖包

sudo apt install libgtk2.0-dev pkg-config

一定要安装,否则会报错

• 缺少gtk

terminate called after throwing an instance of 'cv::Exception'
  what():  OpenCV(4.1.0) /home/lynx/fuck_mount/opencv/source_code/opencv_4_1_0/modules/highgui/src/window.cpp:610: error: (-2:Unspecified error) The function is not implemented. Rebuild the library with Windows, GTK+ 2.x or Cocoa support. If you are on Ubuntu or Debian, install libgtk2.0-dev and pkg-config, then re-run cmake or configure script in function 'cvNamedWindow'

命令行安装(TODO)

1. cmake

cd {生成makefile的位置}
cmake {source的位置}

2. make

cd {生成makefile的位置}
make

3. make install

make install {安装的位置}

gui安装

参考:CSDN

1. 安装cmake-gui

sudo apt install cmake-gui

2. 启动cmake-gui

cd {build二进制文件的目录}
cmake-gui {opencv source code}

3. Configure->Unix Makefiles:Use default native compilers->finish

4. 狂点Configure知道没有红色区域

5. 赋值

6. 狂点Configure

7. Generate,等到Generating done

8. 在build路径make

sudo make -j4 # 具体看有多少cpu
cat /proc/cpuinfo | grep "processor" | wc -l # 查看cpu个数
sudo make install

9. 配置opencv4.pc(2选1)

   • 修改PKG_CONFIG_PATH:增加{你cmake_install的路径}/lib/pkgconfig(opencv4.pc所在的路径)
   • (TODO,没试过)复制opencv4.pc到/usr/lib/pkgconfig

10. 配置库路径,(新建)/etc/ld.so.conf.d/opencv.conf

{你cmake_install路径}/lib/ # 自己的lib路径

sudo ldconfig

11. 编译测试

如果是opencv.pc将opencv4改成opencv

pkg-config --modversion opencv4
g++ a.cpp `pkg-config --cflags --libs opencv4` -o test && ./test

常见问题

1. 在其他项目中用CMake导入OpenCV模块:

   find_package(OpenCV REQUIRED) include_directories(${OpenCV_INCLUDE_DIRS})

之后cmake出现By not providing "FindOpenCV.cmake" in CMAKE_MODULE_PATH ...:

这是由于没有设置环境变量OpenCV_DIR,设置为OpenCV编译后的目录下含有OpenCVConfig.cmake的对应路径