为ARM交叉编译opencv项目并调用go项目gocv

Question

我在主机Ubuntu20.04.3LTS下为ARM编译了opencv和gocv，如下所示。

1. Apt安装arm-linux-gnueabi-gcc，arm-linux-gnueabi-g++

2.编译和安装opencv

cmake -DCMAKE_C_COMPILER=arm-linux-gnueabi-gcc -DCMAKE_CXX_COMPILER=arm-linux-gnueabi-g++  -DCMAKE_TOOLCHAIN_FILE=../opencv/platforms/linux/arm-gnueabi.toolchain.cmake  ../opencv
make -j16
make install

安装后。在当前dir中创建了一个名为install的文件夹，并获得了如下共享库。

../arm2build/install/lib
|-- cmake
|   `-- opencv4
|       |-- OpenCVConfig-version.cmake
|       |-- OpenCVConfig.cmake
|       |-- OpenCVModules-release.cmake
|       `-- OpenCVModules.cmake
|-- libopencv_calib3d.so -> libopencv_calib3d.so.405
|-- libopencv_calib3d.so.4.5.4
|-- libopencv_calib3d.so.405 -> libopencv_calib3d.so.4.5.4
|-- libopencv_core.so -> libopencv_core.so.405
|-- libopencv_core.so.4.5.4
|-- libopencv_core.so.405 -> libopencv_core.so.4.5.4
|-- libopencv_dnn.so -> libopencv_dnn.so.405
|-- libopencv_dnn.so.4.5.4
|-- libopencv_dnn.so.405 -> libopencv_dnn.so.4.5.4
|-- libopencv_features2d.so -> libopencv_features2d.so.405
|-- libopencv_features2d.so.4.5.4
|-- libopencv_features2d.so.405 -> libopencv_features2d.so.4.5.4
|-- libopencv_flann.so -> libopencv_flann.so.405
|-- libopencv_flann.so.4.5.4
|-- libopencv_flann.so.405 -> libopencv_flann.so.4.5.4
|-- libopencv_gapi.so -> libopencv_gapi.so.405
|-- libopencv_gapi.so.4.5.4
|-- libopencv_gapi.so.405 -> libopencv_gapi.so.4.5.4
|-- libopencv_highgui.so -> libopencv_highgui.so.405
|-- libopencv_highgui.so.4.5.4
|-- libopencv_highgui.so.405 -> libopencv_highgui.so.4.5.4
|-- libopencv_imgcodecs.so -> libopencv_imgcodecs.so.405
|-- libopencv_imgcodecs.so.4.5.4
|-- libopencv_imgcodecs.so.405 -> libopencv_imgcodecs.so.4.5.4
|-- libopencv_imgproc.so -> libopencv_imgproc.so.405
|-- libopencv_imgproc.so.4.5.4
|-- libopencv_imgproc.so.405 -> libopencv_imgproc.so.4.5.4
|-- libopencv_ml.so -> libopencv_ml.so.405
|-- libopencv_ml.so.4.5.4
|-- libopencv_ml.so.405 -> libopencv_ml.so.4.5.4
|-- libopencv_objdetect.so -> libopencv_objdetect.so.405
|-- libopencv_objdetect.so.4.5.4
|-- libopencv_objdetect.so.405 -> libopencv_objdetect.so.4.5.4
|-- libopencv_photo.so -> libopencv_photo.so.405
|-- libopencv_photo.so.4.5.4
|-- libopencv_photo.so.405 -> libopencv_photo.so.4.5.4
|-- libopencv_stitching.so -> libopencv_stitching.so.405
|-- libopencv_stitching.so.4.5.4
|-- libopencv_stitching.so.405 -> libopencv_stitching.so.4.5.4
|-- libopencv_video.so -> libopencv_video.so.405
|-- libopencv_video.so.4.5.4
|-- libopencv_video.so.405 -> libopencv_video.so.4.5.4
|-- libopencv_videoio.so -> libopencv_videoio.so.405
|-- libopencv_videoio.so.4.5.4
`-- libopencv_videoio.so.405 -> libopencv_videoio.so.4.5.4

3.用cgo构建gocv/cmd/version项目

CGO_ENABLED=1 CC=arm-linux-gnueabi-gcc CXX=arm-linux-gnueabi-g++ CGO_CPPFLAGS='-I/opt/arm2build/install/include/opencv4' CGO_LDFLAGS='-L/opt/arm2build/lib -lopencv_core -lopencv_videoio -lopencv_imgproc -lopencv_highgui -lopencv_imgcodecs -lopencv_objdetect -lopencv_features2d -lopencv_video -lopencv_dnn -lopencv_calib3d -lopencv_flann -lopencv_photo' GOOS=linux GOARCH=arm go build -tags=customenv .

做完之后会产生ELF

root@793d7120f9df:/opt/go/gocv/cmd/version# file version
version: ELF 32-bit LSB executable, ARM, EABI5 version 1 (SYSV), dynamically linked, interpreter /lib/ld-linux.so.3, BuildID[sha1]=038c80350b7a0c9e72b10021c66c31c2dbb4df2c, for GNU/Linux 3.2.0, not stripped

我的问题是如何使用arm处理器在安卓系统中运行ELF version？

我将install/libs中的静态库复制到移动的/system/lib/中。将version文件复制到/data/local/tmp并授予执行特权。然后执行version得到输出"./version:没有这样的文件或目录“。

130|cactus:/data/local/tmp # ./version
/system/bin/sh: ./version: No such file or directory

Answer 1

我发现原因是动态链接解释器，而android没有linux平台的解释器。用NDK重建opencv后，得到了静态库。用NDK的clang编译器构建go项目，问题就解决了。