CCTag Library

This library provides the code for the detection of CCTag markers made up of concentric circles [CGGG16]. CCTag markers are a robust, highly accurate fiducial system that can be robustly localized in the image even under challenging conditions. The library can efficiently detect the position of the image of the (common) circle center and identify the marker based on the different ratio of their crown sizes.

_images/cctags-example.png

An example of three different CCTag markers with three crowns. Each marker can be uniquely identified thanks to the thickness of each crown, which encodes the information of the marker, typically a unique ID.

The implementation is done in both CPU and GPU (Cuda-enabled cards). The GPU implementation can reach real-time performances on full HD images.

Example of detection in challenging conditions

_images/challenging.png

Examples of synthetic images of circular fiducials under very challenging shooting conditions i.e., perturbed, in particular, by a (unidirectional) motion blur of magnitude 15px. The markers are correctly detected and identified (b,d) with an accuracy of 0.54px and 0.36px resp. in (a) and (c) for the estimated imaged center of the outer ellipse whose semi-major axis (in green) is equal to 31.9px and 34.5px resp.

Comparison with ARToolkitPlus

The video shows the effectiveness and the robustness of the detection compared to the ARToolKitPlus [WS07]. ARTKPlus is, among all the available open-source solutions, one achieving better performances in terms of detection rate and computational time.

In the video, 4 markers for each solution are placed on a plane at known positions, so that the relevant plane-induced homography can be estimated. For ARTKPlus once all the markers are detected and identified, the homography is estimated using all the detected marker corners following a DLT approach [HZ04] (note that the homography can be then estimated even if only one marker is detected). For CCTAG, the image of the four centres of the concentric circles is used to compute the homography.

The image placed in between the markers can be then rectified in order to visually assess the quality of the estimated homography. Thanks to the accurate estimation of the image of the four centres of the concentric circles provided by CCTag, the homography can be robustly estimated and the rectified image is not affected by any significant jittering, whereas the rectified image computed with the ARTKPlus homography is more unstable. Moreover, the video shows that the proposed method allows detecting the marker even in very challenging conditions, such as severe motion blur and sudden illumination changes.

Requirements

Hardware

CCTag has a CPU and a GPU implementation. The GPU implementation requires an NVIDIA GPU card with a CUDA compute capability >= 3.5. You can check your NVIDIA GPU card CC support here or on the NVIDIA dev page. If you do not have a NVIDIA card you will still able to compile and use the CPU version.

Here are the minimum hardware requirements for CCTag:

Minimum requirements

Operating systems

Windows x64, Linux, macOS

CPU

Recent Intel or AMD cpus

RAM Memory

8 GB

Hard Drive

No particular requirements

GPU

NVIDIA CUDA-enabled GPU (compute capability >= 3.5)

Software

CCTag depends on the following libraries:

  • Eigen3 >= 3.3.4

  • Boost >= 1.66

  • OpenCV >= 3.1

  • TBB >= 2021.5.0

Warning

In order to have Cuda support on Windows, at least Eigen 3.3.9 is required


CCTag can be installed from the following package managers.

vcpkg

vcpkg is a cross-platform (Windows, Linux and MacOS), open-source package manager created by Microsoft.

Since v1.0.0 of the library it is possible to build and install the library through vcpkg on Linux, Windows and MacOS by running:

vcpkg install cctag[cuda,apps]

where cuda and apps are the options to build the library with the cuda support and the sample applications, respectively.


conan

conan is a decentralized and multi-platform package manager.

Since v1.0.1, you can install CCTag from conan by running:

conan install cctag/1.0.1@

where 1.0.1@ is the version you want to install. See CCTag as third party for how to use CCTag as third party in your project.


Building the library

Building tools

Required tools:

  • CMake >= 3.14 to build the code

  • Git

  • C/C++ compiler supporting the C++14 standard (gcc >= 5, clang >= 3.4, msvc >= 2017)

Optional tool:

  • CUDA >= 9.0

Note

On Windows, there are compatibility issues to build the GPU part due to conflicts between msvc/nvcc/thrust/eigen/boost.

Dependencies

vcpkg

vcpkg can be used to install all the dependencies on all the supported platforms. This is particularly useful on Windows. To install the dependencies:

vcpkg install
        boost-accumulators
        boost-algorithm
        boost-container
        boost-date-time
        boost-exception
        boost-filesystem
        boost-iterator
        boost-lexical-cast
        boost-math
        boost-mpl
        boost-multi-array
        boost-ptr-container
        boost-program-options
        boost-serialization
        boost-spirit
        boost-static-assert
        boost-stacktrace
        boost-test
        boost-thread
        boost-throw-exception
        boost-timer
        boost-type-traits
        boost-unordered
        opencv
        tbb
        eigen3

You can add the flag --triplet to specify the architecture and the version you want to build. For example:

  • --triplet x64-windows will build the dynamic version for Windows 64 bit

  • --triplet x64-windows-static will build the static version for Windows 64 bit

  • --triplet x64-linux-dynamic will build the dynamic version for Linux 64 bit

and so on. More information can be found here

Linux

On Linux you can install from the package manager:

For Ubuntu/Debian package system:

sudo apt-get install g++ git-all libpng12-dev libjpeg-dev libeigen3-dev libboost-all-dev libtbb-dev

For CentOS package system:

sudo yum install gcc-c++ git libpng-devel libjpeg-turbo-devel eigen3-devel boost-devel      tbb-devel
MacOS

On MacOs using Homebrew install the following packages:

brew install git libpng libjpeg eigen boost tbb

Getting the sources

git clone https://github.com/alicevision/CCTag.git

CMake configuration

From CCTag root folder you can run cmake:

mkdir build && cd build
cmake ..
make -j `nproc`

On Windows add -G "Visual Studio 16 2019" -A x64 to generate the Visual Studio solution according to your VS version (see CMake documentation).

If you are using the dependencies built with VCPKG you need to pass -DCMAKE_TOOLCHAIN_FILE=path/to/vcpkg/scripts/buildsystems/vcpkg.cmake at cmake step to let it know where to find the dependencies.

Otherwise you can specify the path where each dependency can be found (if not installed in system folders) by passing its related path. For example, for OpenCV you can pass -DOpenCV_DIR=path/to/opencv/install/share/OpenCV/ to tell where the OpenCVConfig.cmake file can be found.

CMake options

CMake configuration can be controlled by changing the values of the following variables (here with their default value)

  • CCTAG_WITH_CUDA:BOOL=ON to enable/disable the Cuda implementation

  • BUILD_SHARED_LIBS:BOOL=ON to enable/disable the building shared libraries

  • CCTAG_ENABLE_SIMD_AVX2:BOOL=OFF to enable/disable the AVX2 optimizations

  • CCTAG_BUILD_TESTS:BOOL=OFF to enable/disable the building of the unit tests

  • CCTAG_BUILD_APPS:BOOL=ON to enable/disable the building of applications

  • CCTAG_BUILD_DOC:BOOL=OFF to enable/disable building this documentation

So if you do not want to build the Cuda part, you have to pass -DCCTAG_WITH_CUDA:BOOL=OFF and so on.


CCTag as third party

When you install CCTag a file CCTagConfig.cmake is installed in <install_prefix>/lib/cmake/CCTag/ that allows you to import the library in your CMake project. In your CMakeLists.txt file you can add the dependency in this way:

 1# Find the package from the CCTagConfig.cmake
 2# in <prefix>/lib/cmake/CCTag/. Under the namespace CCTag::
 3# it exposes the target CCTag that allows you to compile
 4# and link with the library
 5find_package(CCTag CONFIG REQUIRED)
 6...
 7# suppose you want to try it out in a executable
 8add_executable(cctagtest yourfile.cpp)
 9# add link to the library
10target_link_libraries(cctagtest PUBLIC CCTag::CCTag)

Then, in order to build just pass the location of CCTagConfig.cmake from the cmake command line:

cmake .. -DCCTag_DIR=$CCTAG_INSTALL/lib/cmake/CCTag/

If you are using conan for your project then you need to add cctag to your conanfile.txt:

[requires]
cctag/1.0.1

[generators]
CMakeToolchain
CMakeDeps

and when building you may need to follow these steps:

mkdir build
cd build
conan install .. -s build_type=Release
cmake .. -DCMAKE_TOOLCHAIN_FILE=conan_toolchain.cmake -DCMAKE_BUILD_TYPE=Release
cmake --build . --config Release

Docker image

A docker image can be built using the Ubuntu based Dockerfile, which is based on nvidia/cuda image (https://hub.docker.com/r/nvidia/cuda/ )

Building the dependency image

We provide a Dockerfile_deps containing a cuda image with all the necessary CCTag dependencies installed.

A parameter CUDA_TAG can be passed when building the image to select the cuda version. Similarly, OS_TAG can be passed to select the Ubuntu version. By default, CUDA_TAG=10.2 and OS_TAG=18.04

For example to create the dependency image based on ubuntu 18.04 with cuda 8.0 for development, use

docker build --build-arg CUDA_TAG=8.0 --tag alicevision/cctag-deps:cuda8.0-ubuntu18.04 -f Dockerfile_deps .

The complete list of available tags can be found on the nvidia dockerhub page

Building the CCTag image

Once you built the dependency image, you can build the cctag image in the same manner using Dockerfile:

docker build --tag alicevision/cctag:cuda8.0-ubuntu18.04 .

Running the CCTag image

In order to run the image nvidia docker is needed: see the installation instruction. Once installed, the docker can be run, e.g., in interactive mode with

docker run -it --runtime=nvidia alicevision/cctag:cuda8.0-ubuntu18.04

Official images on DockeHub

Check the docker hub CCTag repository for the available images.

Library usage

Detecting the markers requires three main entities:

  • the class cctag::CCTag modeling a single marker

  • and the functions cctag::cctagDetection() to process the images and get the list of detected markers.

  • the struc cctag::Parameters that control the detection algorithm through the various parameters that it exposes.

Detection

Here is a minimal sample of code that enable CCTag detection on an image:

 1// set up the parameters
 2const std::size_t nCrowns{3};
 3cctag::Parameters params(nCrowns);
 4// if you want to use GPU
 5params.setUseCuda(true);
 6
 7// load the image e.g. from file
 8cv::Mat src = cv::imread(image_filename);
 9cv::Mat graySrc;
10cv::cvtColor(src, graySrc, CV_BGR2GRAY);
11
12// choose a cuda pipe
13const int pipeId{0};
14
15// an arbitrary id for the frame
16const int frameId{0};
17
18// process the image
19boost::ptr_list<ICCTag> markers{};
20cctagDetection(markers, pipeId, frameId, graySrc, params);

@TODO maybe explain what cuda pipe means

Process detected markers

Here is a simple example on how to process the detected markers. The function drawMarkers takes the list of detected markers and it overlay their information on the original image. From the list of markers, if the detected marker is valid it draws the center of the marker, its ID and the outer ellipse cctag::numerical::geometry::Ellipse, all in green. If the marker is not valid, draw the center and the ID in red.

 1void drawMarkers(const boost::ptr_list<ICCTag>& markers, cv::Mat& image)
 2{
 3    // drawing settings
 4    const int radius{10};
 5    const int fontSize{3};
 6    const int thickness{3};
 7    const int fontFace{cv::FONT_HERSHEY_SIMPLEX};
 8
 9    for(const auto& marker : markers)
10    {
11        // center of the marker
12        const cv::Point center = cv::Point(marker.x(), marker.y());
13        const auto rescaledOuterEllipse = marker.rescaledOuterEllipse();
14
15        // check the status and draw accordingly, green for valid, red otherwise
16        if(marker.getStatus() == status::id_reliable)
17        {
18            const cv::Scalar color = cv::Scalar(0, 255, 0, 255);
19            // draw the center
20            cv::circle(image, center, radius, color, thickness);
21            // write the marker ID
22            cv::putText(image, std::to_string(marker.id()), center, fontFace, fontSize, color, thickness);
23            // draw external ellipse
24            cv::ellipse(image,
25                        center,
26                        cv::Size(rescaledOuterEllipse.a(), rescaledOuterEllipse.b()),
27                        rescaledOuterEllipse.angle() * 180 / boost::math::constants::pi<double>(),
28                        0,
29                        360,
30                        color,
31                        thickness);
32        }
33        else
34        {
35            // the same for invalid markers but in red
36            const cv::Scalar color = cv::Scalar(0, 0, 255, 255);
37            cv::circle(image, center, radius, color, thickness);
38            cv::putText(image, std::to_string(marker.id()), center, fontFace, fontSize, color, thickness);
39        }
40    }
41}

Here is an example of possible result:

_images/cctags-example-detection.png

API References

Main Classes

struct Parameters

Structure containing all the major parameters using in the CCTag detection algorithms.

Public Functions

explicit Parameters(std::size_t nCrowns = kDefaultNCrowns)

The constructor, normally the most interesting parameter is the number of crowns.

Parameters

nCrowns – The number of crowns that the markers to detect are made up of.

template<class Archive>
inline void serialize(Archive &ar, unsigned int version)

Serialize the parameter settings.

Template Parameters

Archive – The class to use to store the data.

Parameters
  • ar[inout] The object where to store the data.

  • version[in] The serialization version.

void setDebugDir(const std::string &debugDir)

Set the debug directory where debug data is stored.

Parameters

debugDir[in]

void setUseCuda(bool val)

Whether to use the Cuda implementation or not.

Note

Ignored if the code is not built with Cuda support.

Parameters

val[in] true to use the Cuda implementation, false to use the CPU.

class CCTag : public cctag::ICCTag

Class modeling the CCTag marker containing the position of the marker in the image, its ID and its status.

Public Functions

inline virtual float x() const override

Ger the x coordinate of the center of the marker.

Returns

x coordinate of the center.

inline virtual float y() const override

Ger the x coordinate of the center of the marker.

Returns

x coordinate of the center.

inline virtual const cctag::numerical::geometry::Ellipse &rescaledOuterEllipse() const override

Get the rescaled outer ellipse of the marker. The rescaled outerEllipse is in the coordinate system of the input image, while the internal ellipse is relative to a pyramid level.

Returns

the outer ellipse.

inline virtual MarkerID id() const override

Get marker ID.

Returns

the marker ID.

inline virtual int getStatus() const override

Get the status of the marker.

Returns

the status of the marker.

Functions

void cctag::cctagDetection(boost::ptr_list<ICCTag> &markers, int pipeId, std::size_t frame, const cv::Mat &graySrc, std::size_t nRings, logtime::Mgmt *durations, const std::string &parameterFilename, const std::string &cctagBankFilename)

Perform the CCTag detection on a gray scale image.

Parameters
  • markers[out] Detected markers. WARNING: only markers with status == 1 are valid ones. (status available via getStatus())

  • pipeId[in] Choose between several CUDA pipeline instances

  • frame[in] A frame number. Can be anything (e.g. 0).

  • graySrc[in] Gray scale input image.

  • nRings[in] Number of CCTag rings.

  • durations[in] Optional object to store execution times.

  • parameterFilename[in] Path to a parameter file. If not provided default parameters will be used.

  • cctagBankFilename[in] Path to the cctag bank. If not provided, radii will be the ones associated to the CCTags contained in the markersToPrint folder.

void cctag::cctagDetection(boost::ptr_list<ICCTag> &markers, int pipeId, std::size_t frame, const cv::Mat &graySrc, const cctag::Parameters &params, logtime::Mgmt *durations = nullptr, const CCTagMarkersBank *pBank = nullptr)

Perform the CCTag detection on a gray scale image.

Parameters
  • markers[out] Detected markers. WARNING: only markers with status == 1 are valid ones. (status available via getStatus())

  • pipeId[in] Choose between several CUDA pipeline instances

  • frame[in] A frame number. Can be anything (e.g. 0).

  • graySrc[in] Gray scale input image.

  • params[in] Parameters for the detection.

  • durations[in] Optional object to store execution times.

  • pBank[in] Path to the cctag bank. If not provided, radii will be the ones associated to the CCTags contained in the markersToPrint folder.

Utility Classes

class Ellipse

It models an ellipse with standard form \( \frac{x^2 - x_c}{a^2} + \frac{y^2 - y_c}{b^2} = 1 \), centered in _center \((x_c, x_y)\) and rotated clock-wise by _angle wrt the x-axis. Note that, arbitrarly, the representation with the major axis aligned with the y-axis is chosen.

Subclassed by cctag::numerical::geometry::Circle

Public Functions

Ellipse() = default

Default constructor, set all parameters to zero.

explicit Ellipse(const Matrix &matrix)

Build an ellipse from a 3x3 matrix representing the ellipse as a conic.

Note

By default, the representation with the major axis aligned with the y-axis is chosen.

Parameters

matrix[in] The 3x3 matrix representing the ellipse.

Ellipse(const Point2d<Eigen::Vector3f> &center, float a, float b, float angle)

Build an ellipse from a set of parameters.

Parameters
  • center[in] The center of the conic.

  • a[in] The length of the semi-axis x.

  • b[in] The length of the semi-axis y.

  • angle[in] The orientation of the ellipse wrt the x-axis as a clock-wise angle in radians.

inline const Matrix &matrix() const

Return the matrix representation of the ellipse.

Returns

3x3 matrix representation of the ellipse.

inline Matrix &matrix()

Return the matrix representation of the ellipse.

Returns

3x3 matrix representation of the ellipse.

inline const Point2d<Eigen::Vector3f> &center() const

Return the center of the ellipse.

Returns

3 element vector with the homogeneous coordinates of the ellipse.

inline Point2d<Eigen::Vector3f> &center()

Return the center of the ellipse.

Returns

3 element vector with the homogeneous coordinates of the ellipse.

inline float a() const

Return the length of the x-semi axis of the ellipse.

Returns

the length of the x-semi axis of the ellipse.

inline float b() const

Return the length of the y-semi axis of the ellipse.

Returns

the length of the y-semi axis of the ellipse.

inline float angle() const

Return the orientation of the ellipse.

Returns

the clock-wise orientation angle in radians of the ellipse wrt the x-axis

void setA(float a)

Set the length of the x-semi axis of the ellipse.

Parameters

a[in] the length of the x-semi axis.

void setB(float b)

Set the length of the y-semi axis of the ellipse.

Parameters

b[in] the length of the y-semi axis.

void setAngle(float angle)

Set the orientation angle of the ellipse.

Parameters

angle[in] the clock-wise orientation angle in radians.

void setCenter(const Point2d<Eigen::Vector3f> &center)

Set the center of the ellipse.

Parameters

center[in] the new center of the ellipse.

void setMatrix(const Matrix &matrix)

Update the ellipse from a matrix representing a conic.

Parameters

matrix[in] 3x3 matric representing the ellipse.

void setParameters(const Point2d<Eigen::Vector3f> &center, float a, float b, float angle)

Update the ellipse from its parameters.

Parameters
  • center[in] The center of the conic.

  • a[in] The length of the semi-axis x.

  • b[in] The length of the semi-axis y.

  • angle[in] The orientation of the ellipse wrt the x-axis as a clock-wise angle in radians.

Ellipse transform(const Matrix &mT) const

Return a new ellipse obtained by applying a transformation to the ellipse.

Parameters

mT[in] a 3x3 matrix representing the transformation.

Returns

the transformed ellipse.

void getCanonicForm(Matrix &mCanonic, Matrix &mTprimal, Matrix &mTdual) const

Compute the canonical form of the conic, along with its transformation.

Parameters
  • mCanonic[out] 3x3 diagonal matrix representing the ellipse in canonical form.

  • mTprimal[out] 3x3 transformation matrix such that C = mTprimal.transpose() * mCanonic * mTprimal

  • mTdual[out] 3x3 inverse transformation matrix (= mTprimal.inv())

Friends

friend std::ostream &operator<<(std::ostream &os, const Ellipse &e)

Print the ellipse in matrix form in Matlab notation.

Parameters
  • os[inout] the stream where to output the ellipse.

  • e[in] the ellipse

Returns

the stream with appended the matrix representation of the ellipse.

struct Mgmt
class Measurement

Markers usage

You can find the pdf of the marker to use in the markersToPrint of the project root directory.

Generate the markers

In the markersToPrint directory you can also find a python program generate.py to generate the svg file of the markers. You can customize the size and print the id of the marker on the corner.

Here is the usage and options:

usage: generate.py [-h] [--rings N] [--outdir dir] [--margin N] [--radius N]
                   [--addId] [--addCross] [--generatePng] [--generatePdf]
                   [--whiteBackground]

Generate the svg file for the markers.

optional arguments:
  -h, --help         show this help message and exit
  --rings N          the number of rings (possible values {3, 4}, default: 3)
  --outdir dir       the directory where to save the files (default: ./)
  --margin N         the margin to add around the external ring (default: 400)
  --radius N         the radius of the outer circle (default: 500)
  --addId            add the marker id on the top left corner
  --addCross         add a small cross in the center of the marker
  --generatePng      also generate a png file
  --generatePdf      also generate a pdf file
  --whiteBackground  set the background (outside the marker) to white instead
                     of transparent

For example, calling:

./generate.py --outdir markers3 --margin 100 --addId

it will create a directory markers3 where it saves an svg file for each marker with a margin around the marker of 100 and with the ID of the marker printed on the top left corner.

To generate pdf and/or png file, use the flags --generatePdf and --generatePng .

About

License

CCTag is licensed under MPLv2 license.

More info about the license and what you can do with the code can be found at tldrlegal website

Contact us

You can contact us on the public mailing list at alicevision@googlegroups.com

You can also contact us privately at alicevision-team@googlegroups.com

Cite us

If you want to cite this work in your publication, please use the following

@inproceedings{calvet2016Detection,
  TITLE = {{Detection and Accurate Localization of Circular Fiducials under Highly Challenging Conditions}},
  AUTHOR = {Calvet, Lilian and Gurdjos, Pierre and Griwodz, Carsten and Gasparini, Simone},
  BOOKTITLE = {{Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}},
  ADDRESS = {Las Vegas, United States},
  PAGES = {562 - 570},
  YEAR = {2016},
  MONTH = Jun,
  DOI = {10.1109/CVPR.2016.67}
}

Acknowledgements

This has been developed in the context of the European project POPART founded by European Union’s Horizon 2020 research and innovation programme under grant agreement No 644874.

Additional contributions for performance optimizations have been funded by the Norwegian RCN FORNY2020 project FLEXCAM.

Bibliography

CGC12

L. Calvet, P. Gurdjos, and V. Charvillat. Camera tracking using concentric circle markers: paradigms and algorithms. In 2012 19th IEEE International Conference on Image Processing. IEEE, September 2012. doi:10.1109/icip.2012.6467121.

CGGG16

Lilian Calvet, Pierre Gurdjos, Carsten Griwodz, and Simone Gasparini. Detection and Accurate Localization of Circular Fiducials under Highly Challenging Conditions. In Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 562 – 570. Las Vegas, United States, June 2016. URL: https://hal.archives-ouvertes.fr/hal-01420665/document, doi:10.1109/CVPR.2016.67.

HZ04

Richard Hartley and Andrew Zisserman. Multiple View Geometry in Computer Vision. Cambridge University Press, second edition, 2004.

WS07

Daniel Wagner and Dieter Schmalstieg. ARToolKitPlus for Pose Tracking on Mobile Devices. In Computer Vision Winter Workshop (CVWW), 139–146. 2007. URL: http://www.icg.tu-graz.ac.at/Members/daniel/ARToolKitPlusMobilePoseTracking.