The article discusses an approach to creating vector geometric models for environment representation. In conjunction these models form a hierarchical map of some area. The problems of a mobile robot's localization and motion control can be solved using these structures. Moreover, vector models provide an additional opportunity for recognizing objects by their geometric characteristics. This is achieved by the invariant features of the model with respect to the motions of the object (robot) in space. A functional feature of the method is processing the secondary data based on a piecewise linear approximation of the edges allocated from the images of the surrounding space. This approach allows to use sensors based on different physical principles of operation. The algorithms of successive refinement of object models and adaptive statistical evaluation of the results are described. These algorithms guarantee the precision of the designed map.