Fiducial Design

From LadypackWiki

Ideally, the robot will be able to use naturally occuring distinctive features in the observed physical space to localize itself and label places and objects. These natural fiducials could be distinctive artwork, desk arrangements, decorations, and office/door labels. This, by itself, is pretty tough to do, and is not the first problem we want to address. For the time being, artificial fiducials will be specially designed to make the localization process much easier.

Requirements

  • detection - fiducials should be easy to detect from a reasonable distance
  • unique id - fiducials should be unique enough that, combined with previous state information, the robot should be able to determine which fiducial it's looking at.
  • pose estimation - fiducials should be designed in such a way that the robot can determine its position and pose relative to a fiducial in its field of fiew

Related work

paper year authors notes
Marker Tracking and HMD Calibration for a Video-based Augmented Reality Conferencing System (http://scholar.google.com/url?sa=U&q=http://www.hitl.washington.edu/artoolkit/Papers/IWAR99.kato.pdf) 1999 H. Kato and M. Billinghurst
A Region Adjacency Tree Approach to the Detection and Design of Fiducials (http://web.media.mit.edu/~enrico/papers/vvg.pdf) 2003 E. Constanza and J. Robinson Binarize image, build graph of connected components. To search for fiducials, search graph of image for a sub-graph that matches a fiducial graph. Slow. Robust to fiducial warping. Not robust against occlusions. No fixed geometry for fiducials (pose estimation much more difficult).
Multi-ring color fiducial systems for scalable fiducial tracking augmented reality (http://scholar.google.com/url?sa=U&q=ftp://ftp.usc.edu/pub/graphics/papers/vrais98.ps.gz) 1998 Y. Cho and U. Neumann multi ring, multi scale color fiducials. don't want/need multi-scale..
What is the best fiducial? (http://metlab.cse.msu.edu/tracking-prepared/charles-owen-art02.pdf) 2002 C. B. Owen and F. Xiao and P. Middlin Compares a few fiducial designs and concludes that the best fiducial is a square border surrounding an interior image constructed from DCT basis functions. Not convinced that the fiducial is easily separated from the rest of the image without many false positives.
TRIP: a Low-Cost Vision-Based Location System for Ubiquitous Computing (http://www-lce.eng.cam.ac.uk/~dl231/PUC/PUCpaper.pdf) 2001 D. Lopez de Ipina
Fitting Parameterized Three-Dimensional Models to Images (http://www.cs.ubc.ca/spider/lowe/papers/pami91.pdf) 1991 D. Lowe
Invariant Descriptors for 3-D Object Recognition and Pose (http://people.csail.mit.edu/albert/reading/vision/pose%20estimation/Invariant%20Descriptors%20for%203-D%20Object%20Recognition%20and%20Pose.pdf) 1991 D. Forsyth and J. L. Mundy and A. Zisserman and C. Coelho and A. Heller and C. Rothwell pose_from_circle
A Versatile Camera Position Measurement System for Virtual Reality TV Production (http://people.csail.mit.edu/albert/reading/vision/pose%20estimation/A%20Versatile%20Camera%20Position%20Measurement%20System%20for%20Virtual%20Reality%20TV%20Production.pdf) 1997 G. A. Thomas and J. Jin and T. Niblett and C. Urquhart markers with lots of little circles
Navigation