Fiducial Marker TrackingDART currently has the ARToolkit embedded into it for marker tracking. We use a slightly modified version, that supports the OpenCV-style camera calibration. We currently store most information (the ARToolkit pattern files, the multi-fiducial panel files, and the camera calibration files) in an external data directory (see documentation for all the various support files on The DART Data Directory page).
Some folks have noticed that tracking flickers. The ARToolkit is very very sensitive to lighting, so try adjusting the threshold values in MarkerTrackingInit. Use can also Use Sub-Pixel corner refinement to improve the marker algorithm. An additional behavior called MarkerDebugControls can be used to debug marker tracking in real-time, using specified keyboard buttons to adjust threshold levels. Finally, adjust the settings for the camera. Many cameras attempt to do automatic adjustments for lighting and color and this might be effecting the performance of marker tracking.
We are working on our own marker tracker, using OpenCV, that may make it into DART soon. We are also planning on incorporating some other computer vision techniques. Note that, now that the webCamXtra supports "pushing" image objects into it instead of needing to use a camera, you could retrieve video from DART cameras (take a look at the behavior that records video to a cast to see how to grab frames) and push them into the webCamXtra to do some simple vision processing. There are probably other Xtras that do these sorts of things too.
Please add any other comments or questions on Marker Tracking or computer vision here. Pointers to other things that do or don't work is welcome.
An obvious choice for another marker tracker is webcamxtra.sourceforge.net/ -oh
Links to this Page