Correspondenceless stereo vision for planar or distant scene

Despite extensive study on it for decades, stereo vision is still regarded by many computer vision researchers as not totally solved. What makes it difficult is the stereo correspondence problem. In 1992 Aloimonos and Herve proposed an algorithm that avoids the need of establishing correspondences across the stereo images. The algorithm however assumes a particular stereo configuration: the parallel-axis stereo geometry. This paper describes how the restriction could be removed and provides, in a closed-form format, an extension of the algorithm for general stereo geometry. Experimental results on real image data are also presented to illustrate the correctness of the derived expression.