Camera auto-calibration from a sequence of images

The estimation of a camera's intrinsic parameters (the camera focal length and the principal point) and extrinsic parameters (camera orientation and pose) is the problem of camera calibration. Camera calibration can be accomplished using one of two strategies. The first is off-line calibration which is often used when the camera is stationary during the setup operation. This however reduces the flexibility of using the camera. Alternatively, the camera may be calibrated during the normal operation of the setup. This strategy is called auto-calibration. To achieve greatest flexibility, this work focuses on the auto-calibration strategy. Once the camera is calibrated, it is possible to compute a metric reconstruction from the image sequence. As a result of the flexibility of this strategy, a 3D scene can be imaged from multiple views by a hand-held camera, which can be moving freely without special setup or hardware configuration. Several methods for solving the camera auto-calibration are discussed and implemented in this thesis. The thesis also studies this problem when the camera internal parameters are allowed to change, as the case with a zooming image sequence.