In the recent past, the MOVI research team developped a number of theories and methods for calibrating one or several cameras. Generally speaking, there are two categories of camera calibration methods. The first category — off-line calibration — uses a calibrated pattern and/or object whose 3-D measurements are precisely known, as well as point correspondences between this object (3-D coordinates) and their image projections (2-D coordinates). The second class of methods (known as self-calibration techniques) uses an unknown rigid object as well as point correspondences between several images of a moving camera or of a moving stereo rig.

The task of calibrating a network of cameras raises several interesting methodological and practical problems. Neither off-line nor self calibration methods are entirely satisfactory. Instead we plan to use a combination of both these two categories. Within the next two years we foresee the development of a multi-camera calibration system. The user will be allowed to place as many cameras as needed at unconstrained locations. The software will include video acquisition, image processing, object tracking, camera calibration, and result validation. It will deliver the intrinsic parameters of each one of the cameras, it’s optical corrections (if any) as well as the Euclidean parameters of the camera setup — the position and orientation of each one of the cameras with respect to a global reference system. These developments will be based on the theoretical and methodological results obtained during the last decade. They are carried out within the SEMOCAP contract.