A CAMERA CALIBRATION TECHNIQUE USING TARGETS OF CIRCULAR FEATURES

Accurate camera calibration is essential in many computer vision applications, where quantitative information is extracted from the images. This paper deals with the problems of subpixel feature location on the calibration target and its automatic matching with the corresponding 3-D world points, as the first part of an unsupervised calibration process. In the proposed method a grid of circular features is used as target. Feature detection and location is carried out using a very simple and efficient connected component labeling algorithm, which incrementally calculates an ellipsoidal description of the regions. This description is a robust and accurate model of the projected circular features, since the perspective projection of a circle is always an ellipse. Some heuristics are presented for selecting those regions corresponding to circles in the target. From these ellipses, feature points are extracted, considering the effect of perspective. Experimental results have shown high robustness of the method against random noise and defocusing. Subpixel accuracy is achieved and computational complexity is linear with the number of pixels.