Vision-guided mobile robot navigation using neural network

Abstract

We propose a new approach to solve the correspondence problem for a set of segments extracted from a pair of stereo images. The problem is first formulated as an optimization task where a cost function, which represents the constraints on the solution, is to be minimized. The optimization problem is then performed by means of a two-dimensional Hopfield neural network. Each image of a pair of stereo images is represented by an adjacency graph to eliminate the possibility of choosing segments that have no chance of being a candidate for a match. To reduce the computation burden of the onboard computer, a system architecture has been developed to provide segment feature information for the stereo correspondence process. Finally, we show numerous results obtained with this approach.