Vision, Modeling, and Visualization (2010) Reinhard Koch, Andreas Kolb, Christof Rezk-Salama (Eds.) CALTag: High Precision Fiducial Markers for Camera Calibration B. Atcheson 1 and F. Heide 2 and W. Heidrich 1 1 University of British Columbia, Canada 2 University of Siegen, Germany Abstract We present a self-identifying marker pattern for camera calibration, together with the associated detection algo- rithm. The pattern is designed to support high-precision, fully-automatic localization of calibration points, as well as identification of individual markers in the presence of significant occlusions, uneven illumination, and observa- tions under extremely acute angles. The detection algorithm is efficient and free of parameters. After calibration we obtain reprojection errors significantly lower than with state-of-the art self-identifying reference patterns. 1. Introduction The typical process for calibrating cameras involves pho- tographing a calibration target from multiple viewpoints, and then identifying calibration points in the image that cor- respond to known points on the target. One of the most frequently-used targets is a black and white planar checker- board, where the calibration points are the corner points be- tween squares. This pattern is simple to produce and allows for high accuracy because the corner points can be detected to subpixel precision [Bou08]. The problem in using checkerboards for camera calibra- tion applications lies in how each corner point is detected and identified. The left and center of Figure 1 show common failure cases for automatic checker detection: partial visibil- ity due to clipping against the image boundary, and due to occlusion. It would be useful if we could just place a scan target directly on top of a calibration pattern for stereo ac- quisition with a handheld camera. This is not possible with checkers due to occlusion and shadows. Instead, the check- ers would have to be geometrically well separated from the scan object, thus reducing both the calibration accuracy and the useful image resolution for the actual target object. Man- ual intervention and labeling can overcome this limitation to some degree, but is cumbersome for multi-camera arrays, videos or large image sequences. An alternative to the common checker board are indi- vidually identifiable (fiducial) markers that allow for detec- tion and thus calibration, even if only a small percentage of Figure 1: Partial visibility due to clipping (left image) or occlusion (center) are common failure points of calibration methods involving a checker pattern. By comparison, a cal- ibration system using fiducial markers such as ours (right) can easily deal with partial visibility. tags are visible. Unfortunately for our purpose, most fiducial markers are designed with AR-style applications in mind, where it is important to create isolated markers at a low spa- tial density. As we will see later, this design compromises the precision of the marker localization. In our work, we fo- cus on the development of a fiducial marker system, which we dub CALTag (“CALibration Tags”) that provides • accurate localization of calibration points using subpixel saddle point finders, • high area density of both calibration points and markers, • robustness under occlusion, uneven illumination, radial distortion and observation under acute angles, • minimization of false positives and false negatives through use of checksums, and • automatic processing without parameter tweaking for convenient handling of videos and large image sequences. c  The Eurographics Association 2010.