Building OpenCV-2.4.X for Freescale's i.MX6 BSP (LTIB)
I was working on this post for a long time, and I didn't share any information before because I Was facing a big problem to get the highgui library successfully built to our embedded system. And as you may know, the highgui is very important when you need to open a video stream, create a camera device and even work with windows in OpenCV.
Last week I managed to build it successfully using the latest Freescale's BSP (kernel 3.0.35) and I'm going to show you how to do that in the following lines.
First of all, you can download the Freescale's i.MX6 BSP here !
Unfortunately getting the latest BSP doesn't mean you are going to get all the latest packages. The only things that are really update in this BSP and probably all others are the kernel and specific drivers for the current hardware you are going to work with. The most commons packages like GTK, GLIB2, PANGO, CAIRO and etc... are not updated, and in order to build the new OpenCV 2.4.X we will need to upgrade these packages and install a new one, the v4l-utils.
So let's stop with this BLAH BLAH BLAH and get into what is really interesting !
Step 1 - Building the dependencies
-----------------------------------------------------------
1) Assuming you already have the BSP installed (min profile) in your host machine, you will need to upgrade the GLIB2 package and install the v2l-utils one.
2) Get the new spec files at: https://github.com/andreluizeng/ltib-pkgs-upgrade
3) the new GLIB2 now has a dependency, the LIBFFI. (you can also get the spec file of this lib in the link above.
4) Build & install the libffi as follows:
mkdir ../ltib/dist/lfs-5.1/libffi
cp ../git/ltib-pkgs-upgrade/libffi.spec ../ltib/dist/lfs-5.1/libffi
cp ../git/ltib-pkgs-upgrade/libffi-3.0.12-includedir-1.patch /opt/freescale/pkgs
cp ../downloads/libffi.tar.gz /opt/freescale/pkgs
* you can find the link that you can get the libffi from.
cd ../ltib
./ltib -p libffi.spec -m prep
./ltib -p libffi.spec -m scbuild
./ltib -p libffi.spec -m scdeploy
5) Build & install the new glib2
cd ../ltib
cp ../git/ltib-pkgs-upgrade/glib2.spec ../ltib/dist/lfs-5.1/glib2
cp ../downloads/glib-2.35.9.tar.xz /opt/freescale/pkgs
./ltib -p glib2.spec -m prep
./ltib -p glib2.spec -m scbuild
./ltib -p glib2.spec -m scdeploy
6) Build & install the v4l-utils packages
mkdir ../ltib/dist/lfs-5.1/v4l-utils
cp ../git/ltib-pkgs-upgrade/v4l-utils.spec ../ltib/dist/lfs-5.1/v4l-utils
cp ../downloads/v4l-utils-0.9.3.tar.bz2 /opt/freescale/pkgs
cd ../ltib
./ltib -p v4l-utils.spec -m prep
./ltib -p v4l-utils.spec -m scbuild
./ltib -p v4l-utils.spec -m scdeploy
After all these steps you should have the necessary dependencies built for now building the OpenCV 2.4.X.
Step 2 - Cross-Compiling OpenCV-2.4.X
-----------------------------------------------------------------
When Cross-compiling using cmake you need a special file containing the information about the toolchain you are going to use. In our case we can create this file as:
1) touch toolchain.cmake & vi toolchain.cmake
2) add the following information:
# this one is important
set( CMAKE_SYSTEM_NAME Linux )
#this one not so much
set( CMAKE_SYSTEM_PROCESSOR arm )
# specify the cross compiler
set( CMAKE_C_COMPILER /opt/freescale/usr/local/gcc-4.6.2-glibc- 2.13-linaro-multilib-2011.12/fsl-linaro-toolchain/bin/arm-none-linux-gnueabi-gcc )
set( CMAKE_CXX_COMPILER /opt/freescale/usr/local/gcc-4.6.2-glibc-2.13-linaro-multilib-2011.12/fsl-linaro-toolchain/bin/arm-none-linux-gnueabi-g++ )
# where is the target environment - point to your rootfs here
set( CMAKE_FIND_ROOT_PATH /home/andre/bsps/imx6x/1301/ltib/rootfs )
# search for programs in the build host directories
set( CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER )
# for libraries and headers in the target directories
set( CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY )
set( CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY )
# point to your rootfs path here
set( CMAKE_CXX_FLAGS "-L/home/andre/bsps/imx6x/1301/ltib/rootfs/usr/lib" )
3) download OpenCV-2.4.X here or here.
4) untar the package:
tar -xvf OpenCV-2.4.2.tar.bz2
5) cd OpenCV-2.4.2
mkdir build
cd build
cp ../../toolchain.cmake .
cmake -DCMAKE_TOOLCHAIN_FILE=toolchain.cmake ../
after these steps you should have the necessary files to be built in your build directory, you now just need to configure which extra packages you want to add or remove for your OpenCV profile.
In your build folder:
6) ccmake .
7) you can let your configuration similar to mine:
BUILD_DOCS OFF
BUILD_EXAMPLES OFF
BUILD_JASPER OFF
BUILD_JPEG OFF
BUILD_PACKAGE ON
BUILD_PERF_TESTS ON
BUILD_PNG OFF
BUILD_SHARED_LIBS ON
BUILD_TESTS ON
BUILD_TIFF OFF
BUILD_WITH_DEBUG_INFO ON
BUILD_ZLIB OFF
BUILD_opencv_calib3d ON
BUILD_opencv_contrib ON
BUILD_opencv_core ON
BUILD_opencv_features2d ON
BUILD_opencv_flann ON
BUILD_opencv_gpu ON
BUILD_opencv_highgui ON
BUILD_opencv_imgproc ON
BUILD_opencv_legacy ON
BUILD_opencv_ml ON
BUILD_opencv_nonfree ON
BUILD_opencv_objdetect ON
BUILD_opencv_photo ON
BUILD_opencv_stitching ON
BUILD_opencv_ts ON
BUILD_opencv_video ON
BUILD_opencv_videostab ON
CMAKE_BUILD_TYPE
CMAKE_CONFIGURATION_TYPES Debug;Release
CMAKE_INSTALL_PREFIX /home/andre/imx_applications/OpenCV-2.4.2/build/install
CMAKE_TOOLCHAIN_FILE /home/andre/imx_applications/OpenCV-2.4.2/toolchain.cmake
CMAKE_VERBOSE OFF
CUDA_BUILD_CUBIN OFF
CUDA_BUILD_EMULATION OFF
CUDA_SDK_ROOT_DIR CUDA_SDK_ROOT_DIR-NOTFOUND
CUDA_TOOLKIT_ROOT_DIR CUDA_TOOLKIT_ROOT_DIR-NOTFOUND
CUDA_VERBOSE_BUILD OFF
EIGEN_INCLUDE_PATH EIGEN_INCLUDE_PATH-NOTFOUND
ENABLE_NOISY_WARNINGS OFF
ENABLE_OMIT_FRAME_POINTER ON
ENABLE_PRECOMPILED_HEADERS ON
ENABLE_PROFILING OFF
ENABLE_SOLUTION_FOLDERS OFF
EXECUTABLE_OUTPUT_PATH /home/andre/imx_applications/OpenCV-2.4.2/build/bin
INSTALL_C_EXAMPLES OFF
INSTALL_PYTHON_EXAMPLES OFF
INSTALL_TO_MANGLED_PATHS OFF
LIBRARY_OUTPUT_PATH_ROOT /home/andre/imx_applications/OpenCV-2.4.2/build
OPENCV_CAN_BREAK_BINARY_COMPAT OFF
OPENCV_CONFIG_FILE_INCLUDE_DIR /home/andre/imx_applications/OpenCV-2.4.2/build
OPENCV_WARNINGS_ARE_ERRORS
OFF OPENEXR_INCLUDE_PATH OPENEXR_INCLUDE_PATH-NOTFOUND PVAPI_INCLUDE_PATH PVAPI_INCLUDE_PATH-NOTFOUND PYTHON_PACKAGES_PATH lib/python2.7/site-packages WITH_1394 ON
WITH_CUBLAS OFF
WITH_CUDA OFF
WITH_CUFFT ON
WITH_EIGEN ON
WITH_FFMPEG ON
WITH_GSTREAMER ON
WITH_GTK ON
WITH_JASPER OFF
WITH_JPEG ON
WITH_OPENEXR ON
WITH_OPENGL OFF
WITH_OPENNI OFF
WITH_PNG ON
WITH_PVAPI ON
WITH_QT OFF
WITH_TBB OFF
WITH_TIFF ON
WITH_UNICAP OFF
WITH_V4L ON
WITH_XIMEA OFF
WITH_XINE OFF
you basically removed the gpu acceleration (we don't support full profile OpenCL), we enabled the v4l library and also disabled some unnecessary ones, you can try your on configuration if you will.
8) make -j
9) make install
If you reached the step 9 with no issues you should have all your cross built OpenCV in the install folder. Just copy it to your rootfs:
10) sudo cp -a build/install ../ltib/rootfs/usr
and you know have a OpenCV-2.4.X installation in your embedded system.
Now the result using a simple code to open a camera device and display the image:
EOF !
I Got some issues building GTK when following the procedure above, so here is the deal to avoid any headache:
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
EOF² !
Last week I managed to build it successfully using the latest Freescale's BSP (kernel 3.0.35) and I'm going to show you how to do that in the following lines.
First of all, you can download the Freescale's i.MX6 BSP here !
Unfortunately getting the latest BSP doesn't mean you are going to get all the latest packages. The only things that are really update in this BSP and probably all others are the kernel and specific drivers for the current hardware you are going to work with. The most commons packages like GTK, GLIB2, PANGO, CAIRO and etc... are not updated, and in order to build the new OpenCV 2.4.X we will need to upgrade these packages and install a new one, the v4l-utils.
So let's stop with this BLAH BLAH BLAH and get into what is really interesting !
Step 1 - Building the dependencies
-----------------------------------------------------------
1) Assuming you already have the BSP installed (min profile) in your host machine, you will need to upgrade the GLIB2 package and install the v2l-utils one.
2) Get the new spec files at: https://github.com/andreluizeng/ltib-pkgs-upgrade
3) the new GLIB2 now has a dependency, the LIBFFI. (you can also get the spec file of this lib in the link above.
4) Build & install the libffi as follows:
mkdir ../ltib/dist/lfs-5.1/libffi
cp ../git/ltib-pkgs-upgrade/libffi.spec ../ltib/dist/lfs-5.1/libffi
cp ../git/ltib-pkgs-upgrade/libffi-3.0.12-includedir-1.patch /opt/freescale/pkgs
cp ../downloads/libffi.tar.gz /opt/freescale/pkgs
* you can find the link that you can get the libffi from.
cd ../ltib
./ltib -p libffi.spec -m prep
./ltib -p libffi.spec -m scbuild
./ltib -p libffi.spec -m scdeploy
5) Build & install the new glib2
cd ../ltib
cp ../git/ltib-pkgs-upgrade/glib2.spec ../ltib/dist/lfs-5.1/glib2
cp ../downloads/glib-2.35.9.tar.xz /opt/freescale/pkgs
./ltib -p glib2.spec -m prep
./ltib -p glib2.spec -m scbuild
./ltib -p glib2.spec -m scdeploy
6) Build & install the v4l-utils packages
mkdir ../ltib/dist/lfs-5.1/v4l-utils
cp ../git/ltib-pkgs-upgrade/v4l-utils.spec ../ltib/dist/lfs-5.1/v4l-utils
cp ../downloads/v4l-utils-0.9.3.tar.bz2 /opt/freescale/pkgs
cd ../ltib
./ltib -p v4l-utils.spec -m prep
./ltib -p v4l-utils.spec -m scbuild
./ltib -p v4l-utils.spec -m scdeploy
After all these steps you should have the necessary dependencies built for now building the OpenCV 2.4.X.
Step 2 - Cross-Compiling OpenCV-2.4.X
-----------------------------------------------------------------
When Cross-compiling using cmake you need a special file containing the information about the toolchain you are going to use. In our case we can create this file as:
1) touch toolchain.cmake & vi toolchain.cmake
2) add the following information:
# this one is important
set( CMAKE_SYSTEM_NAME Linux )
#this one not so much
set( CMAKE_SYSTEM_PROCESSOR arm )
# specify the cross compiler
set( CMAKE_C_COMPILER /opt/freescale/usr/local/gcc-4.6.2-glibc- 2.13-linaro-multilib-2011.12/fsl-linaro-toolchain/bin/arm-none-linux-gnueabi-gcc )
set( CMAKE_CXX_COMPILER /opt/freescale/usr/local/gcc-4.6.2-glibc-2.13-linaro-multilib-2011.12/fsl-linaro-toolchain/bin/arm-none-linux-gnueabi-g++ )
# where is the target environment - point to your rootfs here
set( CMAKE_FIND_ROOT_PATH /home/andre/bsps/imx6x/1301/ltib/rootfs )
# search for programs in the build host directories
set( CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER )
# for libraries and headers in the target directories
set( CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY )
set( CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY )
# point to your rootfs path here
set( CMAKE_CXX_FLAGS "-L/home/andre/bsps/imx6x/1301/ltib/rootfs/usr/lib" )
3) download OpenCV-2.4.X here or here.
4) untar the package:
tar -xvf OpenCV-2.4.2.tar.bz2
5) cd OpenCV-2.4.2
mkdir build
cd build
cp ../../toolchain.cmake .
cmake -DCMAKE_TOOLCHAIN_FILE=toolchain.cmake ../
after these steps you should have the necessary files to be built in your build directory, you now just need to configure which extra packages you want to add or remove for your OpenCV profile.
In your build folder:
6) ccmake .
7) you can let your configuration similar to mine:
BUILD_DOCS OFF
BUILD_EXAMPLES OFF
BUILD_JASPER OFF
BUILD_JPEG OFF
BUILD_PACKAGE ON
BUILD_PERF_TESTS ON
BUILD_PNG OFF
BUILD_SHARED_LIBS ON
BUILD_TESTS ON
BUILD_TIFF OFF
BUILD_WITH_DEBUG_INFO ON
BUILD_ZLIB OFF
BUILD_opencv_calib3d ON
BUILD_opencv_contrib ON
BUILD_opencv_core ON
BUILD_opencv_features2d ON
BUILD_opencv_flann ON
BUILD_opencv_gpu ON
BUILD_opencv_highgui ON
BUILD_opencv_imgproc ON
BUILD_opencv_legacy ON
BUILD_opencv_ml ON
BUILD_opencv_nonfree ON
BUILD_opencv_objdetect ON
BUILD_opencv_photo ON
BUILD_opencv_stitching ON
BUILD_opencv_ts ON
BUILD_opencv_video ON
BUILD_opencv_videostab ON
CMAKE_BUILD_TYPE
CMAKE_CONFIGURATION_TYPES Debug;Release
CMAKE_INSTALL_PREFIX /home/andre/imx_applications/OpenCV-2.4.2/build/install
CMAKE_TOOLCHAIN_FILE /home/andre/imx_applications/OpenCV-2.4.2/toolchain.cmake
CMAKE_VERBOSE OFF
CUDA_BUILD_CUBIN OFF
CUDA_BUILD_EMULATION OFF
CUDA_SDK_ROOT_DIR CUDA_SDK_ROOT_DIR-NOTFOUND
CUDA_TOOLKIT_ROOT_DIR CUDA_TOOLKIT_ROOT_DIR-NOTFOUND
CUDA_VERBOSE_BUILD OFF
EIGEN_INCLUDE_PATH EIGEN_INCLUDE_PATH-NOTFOUND
ENABLE_NOISY_WARNINGS OFF
ENABLE_OMIT_FRAME_POINTER ON
ENABLE_PRECOMPILED_HEADERS ON
ENABLE_PROFILING OFF
ENABLE_SOLUTION_FOLDERS OFF
EXECUTABLE_OUTPUT_PATH /home/andre/imx_applications/OpenCV-2.4.2/build/bin
INSTALL_C_EXAMPLES OFF
INSTALL_PYTHON_EXAMPLES OFF
INSTALL_TO_MANGLED_PATHS OFF
LIBRARY_OUTPUT_PATH_ROOT /home/andre/imx_applications/OpenCV-2.4.2/build
OPENCV_CAN_BREAK_BINARY_COMPAT OFF
OPENCV_CONFIG_FILE_INCLUDE_DIR /home/andre/imx_applications/OpenCV-2.4.2/build
OPENCV_WARNINGS_ARE_ERRORS
OFF OPENEXR_INCLUDE_PATH OPENEXR_INCLUDE_PATH-NOTFOUND PVAPI_INCLUDE_PATH PVAPI_INCLUDE_PATH-NOTFOUND PYTHON_PACKAGES_PATH lib/python2.7/site-packages WITH_1394 ON
WITH_CUBLAS OFF
WITH_CUDA OFF
WITH_CUFFT ON
WITH_EIGEN ON
WITH_FFMPEG ON
WITH_GSTREAMER ON
WITH_GTK ON
WITH_JASPER OFF
WITH_JPEG ON
WITH_OPENEXR ON
WITH_OPENGL OFF
WITH_OPENNI OFF
WITH_PNG ON
WITH_PVAPI ON
WITH_QT OFF
WITH_TBB OFF
WITH_TIFF ON
WITH_UNICAP OFF
WITH_V4L ON
WITH_XIMEA OFF
WITH_XINE OFF
you basically removed the gpu acceleration (we don't support full profile OpenCL), we enabled the v4l library and also disabled some unnecessary ones, you can try your on configuration if you will.
8) make -j
9) make install
If you reached the step 9 with no issues you should have all your cross built OpenCV in the install folder. Just copy it to your rootfs:
10) sudo cp -a build/install ../ltib/rootfs/usr
and you know have a OpenCV-2.4.X installation in your embedded system.
Now the result using a simple code to open a camera device and display the image:
EOF !
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
UPDATE
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
I Got some issues building GTK when following the procedure above, so here is the deal to avoid any headache:
- Install X11 (package list)
- Install GTK+ (package list)
- add the LIBFFI and V4L-UTILS like described earlier.
- Upgrade the GLIB2 package like described earlier.
- Cross Compile the OpenCV Library.
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
EOF² !