This paper presents a mobile robot localization approach aimed at using visual information to improve the dead reckoning pose estimates. Visual features are obtained by means of an Inverse Perspective Transformation method, which is able to distinguish between ground points and obstacles. Thanks to this, ground points can be labelled with unique identifiers and their coordinates can be computed. These labelled ground points constitute the visual landmarks and are fused with dead reckoning information by means of an Extended Kalman Filter. The experimental results provided emphasize the benefits of this approach.