We propose a new approach for stereo matching in Autonomous Mobile Robot applications. In this framework an accurate but slow reconstruction of the 3D scene is not needed; rather, it is more important to have a fast localization of the obstacles to avoid them. All the methods in the literature are based on a punctual correspondence, but they are inefficient in realistic contexts for the presence of uniform patterns, or some perturbations between the two images of the stereo pair. Our idea is to face the stereo matching problem as a matching between homologous regions, instead of a point matching. The stereo images are represented as graphs and a graph matching is computed to find homologous regions. We present some results on a standard stereo database and also on a more realistic stereo sequence acquired from a robot moving in an indoor environment, and a performance comparison with other approaches in the literature is reported and discussed. Our method is strongly robust in case of some fluctuations of the stereo pair, homogeneous and repetitive regions, and is fast. The result is a semi-dense disparity map, leaving only a few regions in the scene unmatched