While solved in classical contexts, camera calibration (i.e. recovering intrinsic and extrinsic characteristics of cameras) is still an open problem when considering multiple camera environments without reference points and with moving cameras. This is mainly due to the fact that matching points between wide-baseline camera pairs is an inherently difficult problem.

Our objectives are therefore to study new self-calibration techniques based on silhouettes (image regions where objects or persons project). They will be very useful for automatically updating the calibration of any environment equipped with several cameras, as well as for calibrating a moving camera in such environments; a person or an object present in the environment will be sufficient for these tasks. As mentioned earlier, although self-calibration techniques based on points or segments are well known, the theory still needs to be completed when silhouettes are considered.