Crossing an intersection is a challenge for visually impaired people. While tactile maps can be a medium for appropriating this complex space, they benefit from being complemented by audio information. In this paper we propose a data model to describe an intersection, the paths that allow to cross it, and their accessibility attributes. We also present methods to generate this model automatically from OpenStreetMap, by inferring missing data through graph analysis techniques. Finally, we present an implementation, the evaluation of which confirms the ability of the model to generate a compliant description for intersections with enough data.