Image Component Library (ICL)
FiducialDetector plugins


struct  icl::markers::FiducialDetectorPlugin
 Generic Interface class for FiducialDetector plugins. More...
class  icl::markers::FiducialDetectorPluginAmoeba
 FiducialDetectorPlugin for reacTIVision's 'amoeba' markers. More...
class  icl::markers::FiducialDetectorPluginART
 FiducialDetectorPlugin for ARToolkit like markers using binary image patches as marker IDs. More...
class  icl::markers::FiducialDetectorPluginBCH
 FiducialDetectorPlugin for ARToolkit+ like markers using BCH coded IDs. More...
class  icl::markers::FiducialDetectorPluginForQuads
 FiducialDetectorPlugin for quad-markers like ARToolkit and BCH-Code markers. More...
class  icl::markers::FiducialDetectorPluginHierarchical
 Extra abstraction layer that defines a basic skeleton for the detection of hierarchical fiducials. More...
class  icl::markers::FiducialDetectorPluginICL1
 FiducialDetectorPlugin for ARToolkit+ like markers using BCH coded IDs. More...

Detailed Description

The ICLMarkers package defines general interfaces for most different types of marker detection methods. The essential classes of this package are

Note: We use the words "fiducial" and "marker" or "image marker" as synonyms.

The FiducialDetector implements a plugin-based system which can be configured at instantiation type in order to use a certain marker detection backend. Dependent on the selected backend, the detection markers (of type icl::Fiducial) provide a different set of information -- some can only provide 2D information, others do also provide 3D pose information. Also dependent on the chosen plugin type, markers have to be loaded in a certain way that is also generalized by single method FiducialDetector::loadMarkers.


Internally, most detection steps are optimized for beeing fast on large images (>= VGA resolution),i.e. usually the plugins try to get away from the pixel representation by switching to binarized image regions. Benchmarks for different types of input image sizes and marker types are given in the Benchmarks section

Supported Marker Types

"art" marker "bch" marker "amoeba" marker "icl1" marker




We did not speed up the algorithms using multithreading techniques since usually the provided methods are fast enough using a single thread. "Fast enough" means, that the detection is faster than the amount of data, that is usually provided by common cameras. Perhaps, multithreading will added later as a 'utils::Configurable' property.


#include <ICLQt/Common.h>
#include <ICLMarkers/FiducialDetector.h>

// static application data
HSplit hsplit;
GUI gui(hsplit);
GenericGrabber grabber;

// the global detector class
// here, using the first 100 "bch"-markers
icl::markers::FiducialDetector fid("bch","[0-100]","size=30x30");

// initialization function (called once after start)
void init(){
  // the the configurable ID

  // create the GUI
  gui << Draw().handle("draw")        // create drawing component
      << Prop("fid").maxSize(18,100)  // create the propery widged for 'fid'
      << Show();                      // show the main widget

  // initialize the grabber from given program argument

// working loop (automatically looped in the working thread)
void run(){
  // get a handle to the "draw" component
  static DrawHandle draw = gui["draw"];

  // grab the next image
  const core::ImgBase *image = grabber.grab();

  // detect markers
  const std::vector<Fiducial> &fids = fid.detect(image);

  // visualize the image
  draw = image;

  // draw marker detection results
  for(unsigned int i=0;i<fids.size();++i){
    utils::Point32f c = fids[i].getCenter2D();
    float rot = fids[i].getRotation2D();

    draw->line(c,c+utils::Point32f( cos(rot), sin(rot))*100 );


  // update the visualization

// main method
int main(int n, char **ppc){
  return ICLApp(n,ppc,"[m]-input|-i(2)",init,run).exec();
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Defines