In order to improve the performance of estimating the fundamental matrix, a key problem arising in stereo vision, a novel method based on stripe constraints is presented. In contrast to traditional methods based on algebraic least-square algorithms, the proposed approach aims to minimize a cost function that is derived from the minimum radius of the Hough transform. In a structured-light system with a particular stripe code pattern, there are linear constraints that the points with the same code are on the same surface. Using the Hough transform, the pixels with the same code map to the Hough space, and the radius of the intersections can be defined as the evaluation function in the optimization progress. The global optimum solution of the fundamental matrix can be estimated using a Levenberg- Marquardt optimization iterative process based on the Hough transform radius. Results illustrate the validity of this algorithm, and prove that this method can obtain good performance with high efficiency.
