A triangulation-like solution to the localization problem is described: as opposed to classical triangulation, only two landmarks need to be segmented and matched rather than three. This is obtained at the expense of estimating the distances to the landmarks besides their bearings. The reduced amount of necessary segmentation and matching operations (intrinsically poorly robust) is a fundamental advantage in highly dynamic environments. It is shown how, under mild hypothesis that are commonly satisfied in partially structured environments, distances may be estimated by using an omnidirectional vision system. Robustness and accuracy performances are thoroughly analysed and experimental results are reported