Stereo correspondence through multiple constraint neural networks

Abstract

A fast and robust artificial neural network algorithm for solving the stereo correspondence problem in binocular vision is described. The stereo correspondence is modeled as a cost minimization problem where the cost is the value of the matching function between the edge pixels along the same epipolar line. A multiple-constraint energy minimization neural network is implemented for this matching process. This algorithm integrates ordering and geometry constraints in addition to uniqueness, continuity, and epipolar line constraints into a neural network framework. The interconnections of the neural network are designed to favor the match selections that follow the ordering constraint. The uniqueness, continuity, epipolar line, and geometry constraints are achieved by exciting and inhibiting neurons, and by injecting external bias currents. The result from the neural network is the solution incorporating simultaneously all the above multiple constraints. Various complexity real images are tested to show the robustness. The algorithm is discussed in detail and experimental results using real images are presented. >