Tracking camera calibration in multi-camera sequences through automatic feature detection and matching

When using multi-camera acquisition systems, in order to extract accurate 3D information on the viewed scene, the geometrical, optical and electric characteristics of the camera system must be known with good accuracy. If the acquisition of the sequence of multi-views requires a certain time to be completed, mechanical shocks, vibrations or thermal effects on the cameras and their support can cause a drift of the initial camera parameters. In this paper we propose a technique that, during the acquisition session, is able to track the camera parameters and, whenever possible, to correct them according to the occurred modifications. This technique does not need any a-priori knowledge or test objects to be placed in the scene, but exploits luminance features already present in the scene, such as luminance corners and spots.