Click here for full text:
A New Scale Arbitration Algorithm for Image Sequences Applied to Cylindrical Photographs
Livingston, Mark A.; Sobel, Irwin E.
Keyword(s): camera calibration; colorimetry
Abstract: We chose cylindrical cameras for our work in reconstructing a scene that surrounds the camera positions. While previous algorithms for calibration exist, we found that improvements were needed to give reasonable results for our data, especially in the determination of the relative scale of translations between the cylinders. One approach to determining camera motion is from corresponding elements in the images, for example points or lines. One class of solutions is based on finding a transformation that yields coplanarity of certain vectors. This type of solution is not limited to any particular form of camera. This algorithm adapts to omnidirectional imagers with minor modifications in the constraint matrices. But the algorithm can only determine five degrees of freedom. A scale arbitration algorithm must determine the sixth degree of freedom, the relative lengths of the translation. While robust algorithms have been proposed for determining the relative scale, we found that standard robust statistics were insufficient for registering our cylindrical camera locations. Thus we introduce a new scale arbitration algorithm that takes into account the confidence with which a point has been triangulated using a previous camera pair. It then decides how to use the point to determine scaling for future camera pairs. This issue arises in standard rectilinear images as well, and our confidence metric adapts to any imager and other solution methods. We also address consistent illumination of cylindrical images and an ambiguity that can arise in the rotation and translation direction computations.
Back to Index