Auto-calibration of Pan-Tilt Cameras including Radial Distortion and Zoom Ricardo Galego, Alexandre Bernardino, and Jos´ e Gaspar Institute for Systems and Robotics, Instituto Superior T´ ecnico / UTL, Lisboa, Portugal {rgalego,alex,jag}@isr.ist.utl.pt Abstract. Although there are intrinsic advantages of using pan-tilt-zoom cam- eras their application in automatic surveillance systems is still scarce. The dif- ficulty of creating background models for moving cameras and the difficulty of keeping fitted pose and optical geometrical projection models are key reasons for the limited use of pan-tilt-zoom cameras. Geometric calibration is a useful tool to overcome these difficulties. In this work we propose a method for PTZ camera auto-calibration over the cam- era’s zoom range. A new approach based on a division distortion model, which allows designing linear algorithms, is followed to solve the radial distortion when images are captured in different zooms. Results obtained over both synthetic and real data show that a full zoom range, complete field of view, pan-tilt-zoom cam- era calibration is possible. 1 Introduction Despite the high versatility and potential of pan-tilt-zoom cameras, their use in current surveillance systems is still much less frequent than the use of fixed cameras. The lim- ited dissemination of pan-tilt-zoom cameras can be justified by the slightly higher costs of the hardware and the higher risk of failure due to the mechanical components. This is however just a partial justification, since the hardware costs and failures can decrease significantly with mass production. Other, more compelling, justifications arise from the operation of the cameras and the difficulty of developing surveillance methodologies. Aiming to have mostly high quality employment implies that installations with numer- ous pan-tilt-zoom cameras cannot involve many human operators. This motivates devel- oping automatic control and surveillance methodologies in opposition to the currently utilized manual control. The difficulty of creating background models for moving cam- eras and the difficulty of keeping fitted pose and optical geometrical projection models are also key reasons for the limited availability of automatic surveillance methodolo- gies provided by the industry. Geometric calibration is a useful tool to overcome these difficulties. There are several methods documented in the literature for calibrating cameras. The method proposed by Bouguet [1] is nowadays one of the most used calibration methods. Bouguet’s method allows estimating intrinsic and radial distortion parameters based on imaging a planar chess pattern placed at various orientations. In the case of mobile cameras, in particular the pan-tilt-zoom cameras, there has been shown that they can be auto-calibrated using natural features of a static scenario. Hartley [2] presented an International Symposium on Visual Computing (ISVC), July 16-18, 2012