Exploiting Geometric Restrictions in a PTZ Camera for Finding Point-orrespondences Between Configurations

Abstract

A pan-tilt-zoom (PTZ) camera, fixed in location, mayperform only rotational movements. There is a class offeature-based self-calibration approaches that exploit therestrictions on the camera motion in order to obtain accuratepoint-correspondences between two configurations ofa PTZ camera. Most of these approaches require extensivecomputation and yet do not guarantee a satisfactory result.In this paper, we approach this problem from a differentperspective. We exploit the geometric restrictions on the imageplanes, which are imposed by the motion restrictions onthe camera. We present a simple method for estimating thecamera focal length and finding the point-correspondencesbetween two camera configurations. We compute pan-only,tilt-only and zoom-only correspondences and then combinethe three to derive the geometrical relationship between anytwo camera configurations. We perform radial lens distortionestimation in order to calibrate distorted image coordinates.Our purely geometric approach does not require anyintensive computations, feature tracking or training. However,our point-correspondence experiments show that, itstill performs well-enough for most computer vision applicationsof PTZ cameras.