ICL can be downloaded as source code via SVN or as binary debian packages (soon). Please refer to the download instructions for details. ICL uses standard CMake as its build system (for more details on CMake visit http://www.cmake.org/). ICL comes with only very few installation dependencies. We decided to make a small set of dependencies compulsory in order to limit the set of possible combinations.
Most external dependencies are kept purely optional. The few mandatory dependencies are very general libraries already installed on most systems. For the debian-package based install, dependencies are automatically installed recursively by the package manager. However, the debian package-based installation does only include dependencies that are available as standard debian packages.
In case of compiling ICL manually from source, the -dev packages that include C/C++-headers are needed. The corresponding Ubuntu packages can be installed via:
sudo apt-get install libjpeg-dev libpng-dev libz-dev
The libpthread-dev library comes with the C/C++ compiler
In general each optional dependency adds some extra functionality to ICL. In some of the cases, also a slower but native fallback implementation is provided. In the following, the external dependencies are listed and their benefits are explained.
The Intel IPP is a proprietary library that provides a very large set of highly optimized functions for different domains, such as linear algebra and in particular computer-vision and image processing. Important: Since Intel IPP is proprietary software, Intel IPP linkage must be established by manually compiling ICL from source. However, we plan to replace the static linkage against Intel IPP with a run-time linking approach that would also work for binary installation sources.
The Intel Math Kernel Library dependency is quite similar to the Intel IPP dependency. However, Intel MKL is only used for a much smaller set of linear algebra functions. Global mathematical utility functions such as math::big_matrix_mult_t or DynMatrix::big_matrix_pinv significantly accelerated if Intel MKL is available. However, in contrast to the Intel IPP dependency, all MKL-accelerated functions have an equivalent C++ fallback implementation.
We use OpenCV mainly in order to provide a compatibility interface that converts OpenCV’s common image data types IplImage and CvMat into ICL’s images type core::ImgBase and vice versa. the header OpenCV.h in the ICLCore module provides efficient and simple to use converter methods. These are only available if the OpenCV dependency is met. In addition, OpenCV is currently needed for the LibOpenSurf (http://www.chrisevansdev.com/computer-vision-opensurf.html) based backend of the cv::SurfFeatureDetector which is directly part of ICL.
LibMesaSR is a proprietary library that allows to grab images from SwissRanger 3D time-of-flight cameras provided by the Mesa Imaging company (http://www.mesa-imaging.ch) The library is closed source.
ImageMagick is used to provide a large set of support image types. Most types are supported in both reading and writing. Without ImageMagick, only a few image data types are supported: .ppm, .pnm and .pgm as well as ICL’s internal image format .bicl are natively supported, .png and .jpg are supported explicitly by other external dependencies.
The dc1394 (digital firewire camera) library allows to grab image from firewire cameras and to set camera parameters.
The libfreenect provides a lightweight interface for grabbing images from Microsoft Kinect cameras. The library allows to grab color, depth and IR-images and to set some internal camera properties.
The xine library provides a very intuitive yet powerful interface for grabbing video in a frame-by-frame manner.
The well known Qt Library is used for ICL’s rapid GUI creation toolkit. Actually Qt is also a prerequisite for most ICL applications and for the whole ICLQt module. We strongly recommend to have at least Qt support when building ICL. The Qt package right now also used the OpenGL extension wrangler library libglew-dev and it needs OpenGL headers to be installed.
The closed source basler pylon drivers (including the Genicam libraries) are used for accessing Gigabit-Ethernet (GIG-E) cameras.
(see also Installing and Using Basler Pylon Drivers)
OpenCL is used to significantly speed up a set of processing units using the computing units of graphics cards or other OpenCL platforms. We mainly use it for point cloud processing units located in the ICLGeom module.
In order to build ICL’s API reference, doxygen needs to be installed. Since also generation of inheritance graphs is activated, also ‘dot’ is needed. On Ubuntu, you can install these dependencies using:
sudo apt-get install doxygen graphviz
This will generate the target ‘api’ in the build directory.
In additionn to the API reference, the sphinx-based manual can be build. To this end, you’ll need to have the API dependencies plus the sphinx-build tool, which can be installed on Ubuntu systems using:
sudo apt-get install python-sphinx python-setuptools
sudo easy_install pyparsing
After this, configuring ICL will also create a ‘manual’ target which generates the ICL manual in html-form. Both, manual and api can be triggered by typing:
make doc
from the build directory
Right now, we only use OpenNI as an alternative backend to grab images from Kinect and other PrimeSense 3D cameras
(see also Installing OpenNI / Nite)
PCL has become some kind of a quasi-standard for point cloud processing. ICL’s ICLGeom module provides the generic geom::PointCloudObjectBase interface that is implemented by the geom::PCLPointCloudObject class. In case of having PCL support, ICL can seamlessly interface to PCL algorithms using this class.
Todo
There is an unsolved dependency between PCL and OpenNI, since our PCD-File Grabber uses libpcl-io, which in turn depends on openni.
The robotics service bus is a new and versatile library for interprocess communications. ICL uses it as backends for the io::GenericGrabber and the io::GenericImageOutput to exchange image data between different processes and PCs.
ICL uses CMake as build system. After checking out the sources, it is recommended to used an extra build folder in order keep the source tree clear of any build artefacts:
svn co https://opensource.cit-ec.de/svn/icl/trunk ICL
cd ICL
mkdir build
cd build
Now you can either use cmake’s Qt-gui to configure and to generate the build-system:
cmake-gui ..
Or you can configure your ICL-build from command line using the cmake command. Each dependency XXX can manually be activated by adding a -DBUILD_WITH_XXX=TRUE command line option; by default, all dependencies are deactivated. If dependencies are not to be found in the system’s default directories (e.g. not in /usr or in /) its root directory can be specified by also adding a command line token -DXXX_ROOT=/foo/bar. Please note that adding a root only does not activate the dependency. Some dependencies, such as OpenCV or PCL, also provide an own FindXXX.cmake file, which is usually located in SOMEWHERE/share/XXX. If this is to be used, a token -DXXX_DIR=SOMEWHERE/share/XXX has to be passed instead of the -DXXX_ROOT one. A list of supported dependencies can be obtained by calling:
cmake .. > /dev/null && cmake -L .. | grep BUILD_WITH
from the build folder, which will configure ICL without any dependencies before getting a list of the dependencies. The dependency-less configuration is automatically overwritten by later cmake-calls.
In addition to the definition of dependencies and their root folder, further options are interesting, in particular, the installation prefix, which is set in cmake default manner by adding a -DCMAKE_INSTALL_PREFIX=SOMEWHERE token to the command line options. The build-type (release of debug) can be specified by adding -DCMAKE_BUILD_TYPE=Release|Debug, where by default, release is used. A release build will automatically set the optimization level to -O3 and logically switch off debugging symbols. Debug switches off optimizations using -O0 and enables full debugging symbols -g3.
Further optimizations can be manually enabled. these can be listed by using:
cmake .. > /dev/null && cmake -L .. | grep ENABLE_
Right now, this is:
-DENABLE_FASTMATH_BUILD_OPTION=ON|OFF
-DENABLE_NATIVE_BUILD_OPTION=ON|OFF
-DENABLE_OPENMP_BUILD_OPTION=ON|OFF
-DENABLE_SSEFPMATH_BUILD_OPTION=ON|OFF
Lastly one can define, whether applications, examples and demos are also compiled and installed. Here the options:
-DBUILD_EXAMPLES=ON|OFF
-DBUILD_APPS=ON|OFF
-DBUILD_DEMOS=ON|OFF
have to be used. A demo bash-script that enables some dependencies and defines some variables can be found SOURCE_ROOT/scripts/compileICL.sh. Here is an example
cmake -DBUILD_WITH_IPP=TRUE -DIPP_ROOT=/vol/nivision/share/IPP/7.06 \
-DBUILD_WITH_MKL=TRUE -DMKL_ROOT=/vol/nivision/share/MKL/10.3.11 \
-DBUILD_WITH_EIGEN3=TRUE \
-DBUILD_WITH_V4L=TRUE \
-DBUILD_WITH_XINE=TRUE \
-DBUILD_WITH_LIBFREENECT=TRUE \
-DBUILD_WITH_QT=TRUE \
-DBUILD_WITH_LIBDC=TRUE \
-DBUILD_WITH_OPENCL=TRUE \
-DBUILD_WITH_OPENCV=TRUE -DOpenCV_DIR=/usr/share/OpenCV \
-DBUILD_WITH_IMAGEMAGICK=TRUE \
-DBUILD_WITH_PCL=FALSE -DPCL_DIR=/usr/local/share/pcl-1.6 \
-DBUILD_EXAMPLES=ON \
-DBUILD_DEMOS=ON \
-DBUILD_APPS=ON \
-DCMAKE_INSTALL_PREFIX=/vol/nivision/ \
-DCMAKE_BUILD_TYPE=Release \
-DENABLE_OPENMP_BUILD_OPTION=ON \
-DENABLE_NATIVE_BUILD_OPTION=ON \
-DENABLE_SSEFPMATH_BUILD_OPTION=ON \
-DENABLE_FASTMATH_BUILD_OPTION=ON \
..
Binary packages are not yet supported, but we plan to support this as soon as possible.
Since some special dependencies are more difficult to get running, we will here share our experiences with you.
Download binary packages e.g.
For developing and running applications with pylon two environment variables must be exported:
export PYLON_ROOT=/your-desired-install-directory
export GENICAM_ROOT_V2_1=${PYLON_ROOT}/genicam
To extract and install pylon type:
tar -xzf pylon-2.3.3-1337-32.tar.gz # ...-62-tar-gz for 64-bit version
cd pylon-2.3.3-1337-bininst/
mkdir $(PYLON_ROOT) # if not already existing
tar -C $(PYLON_ROOT) -xzf pylon-bininst-32.tar.gz # ...-62-tar-gz for 64-bit version
Because of the way Pylon is using shared libraries it may be, that some libraries form the Pylon distribution can not be found at runtime, although the corresponding path is provided in the rpath-list. In this case it is necessary to add the following library search path.:
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:${PYLON_ROOT}/lib
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:${PYLON_ROOT}/genicam/bin/Linux32_i86
or for 64-bit version:
export LD_LIBRARY_PATH=${PYLON_ROOT}/lib64
Further suggestions:
To check whether pylon can establish a connection to a camera the IpConfigurator can be used.:
export PATH=${PATH}:${PYLON_ROOT}/bin
IpConfigurator
When the IpConfigurator does not find the camera, Pylon and accordingly ICL will neither. In that case the camera is most likely not in the same ip-address block. Unfortunately it is not possible to change the cameras ip-settings without the IpConfigurator. There are two known workarounds in this case. Setting the ip-address of the computer to the same address block as the camera or using the Windows version of the IpConfigurator - which does not seem to have this problem - to change the cameras ip settings once to the correct block.
Todo
write how to install use OpenNI and Nite