Active Self-calibration of Multi-camera Systems Marcel Br¨ uckner and Joachim Denzler Chair for Computer Vision Friedrich Schiller University of Jena {marcel.brueckner, joachim.denzler}@uni-jena.de Abstract. We present a method for actively calibrating a multi-camera system consisting of pan-tilt zoom cameras. After a coarse initial cali- bration, we determine the probability of each relative pose using a prob- ability distribution based on the camera images. The relative poses are optimized by rotating and zooming each camera pair in a way that sig- nificantly simplifies the problem of extracting correct point correspon- dences. In a final step we use active camera control, the optimized rela- tive poses, and their probabilities to calibrate the complete multi-camera system with a minimal number of relative poses. During this process we estimate the translation scales in a camera triangle using only two of the three relative poses and no point correspondences. Quantitative experi- ments on real data outline the robustness and accuracy of our approach. 1 Introduction In the recent years multi-camera systems became increasingly important in com- puter vision. Many applications take advantage of multiple cameras observing a scene. Multi-camera systems become even more powerful if they consist of active cameras, i.e. pan-tilt zoom cameras (Fig. 1). For many applications, however, the (active) multi-camera system needs to be calibrated, i. e. the intrinsic and extrinsic parameters of the cameras have to be determined. Intrinsic param- eters of a camera can be estimated using a calibration pattern [1] or camera self-calibration methods for a rotating camera [2, 3]. The focus of this paper is on (active) extrinsic calibration which consists of estimating the rotation and translation of each camera relative to some common world coordinate system. Classical methods for extrinsic multi-camera calibration need a special cali- bration pattern [1] or user interaction like a moving LED in a dark room [4, 5]. From a practical point of view, however, a pure self-calibration is most appeal- ing. Self-calibration in this context means that no artificial landmarks or user interaction are necessary. The cameras estimate their position only from the im- ages they record. An example for self-calibration of a static multi-camera system is the work of L¨ abe and F¨ orstner [6]. Given several images they extract point correspondences and use these to estimate the relative poses. Another example is the graph based calibration method proposed by Bajramovic and Denzler [7] which considers the uncertainty of the estimated relative pose of each camera pair. However, both methods are designed for static cameras and do not use the benefits of active camera control.