A method with low computational cost is proposed to solve the map building and localization problems. Based on omni-vision which can provide rich environment information, this method constructs a descriptor for each environment point. Then a topology map is made as a set of environment descriptors. In the meantime, a Bayesian based localization algorithm, which uses such omni-vision based topology map, is presented. This algorithm can solve different kinds of localization problems, even the hardest “kidnapping” localization problem. The experimental results show the effectiveness and feasibility of the proposed method.