This paper presents a vision based localization system which can be used on a wide range of mobile vehicles. The principle of the algorithm developed is to merge the information given by a vision system with the data extracted from the moves of the vehicle. The image processing level is performed by using principal components analysis that allows low cost position estimation by using a representative set of images obtained during an initial exploration of the environment. This set of images makes it possible to represent the environment as a partially observable Markov decision process. The originality of this approach is the resulting data fusion process that uses both image matching and decision made by the robot in order to estimate the set of plausible positions of the robot and the associated probabilities. Furthermore, this stochastic localisation device shows better results compared with the classical methods such as static neighborhoods. The main characteristics of this localization device are its robustness, its accuracy and its low cost compared with usual localization methods.