Abstract: The acquisition of 3-D scenes by a single hand-held camera is disclosed. The camera is preferably mounted on a lens attachment with four mirrors, enabling stereo capturing of the scene. The signal processing tasks involved, camera calibration and correspondence estimation, are each disclosed. Both fixed and self-calibration methods benefit from the use of the lens attachment to acquire a scene. Specifically, full self-calibration of the camera is enabled, without loss of absolute scale as seen in prior art stereo self-calibration methods. Accurate 3-D models are obtained with the method of the invention.