Traffic Sign Recognition (TSR) system is an important component for the intelligent vehicles, it can assist and inform the driver about dangerous situations such as stop, icy roads, no entry or speed limits. In this paper we present a fast and robust traffic sign recognition system constituted of three modules which are: segmentation, detection and recognition of sign type. In the first module we start by applying a filter after normalization of the three RGB channels to extract red, green, blue and yellow maps. To detect the signs and identify their forms, in the second module we propose a new and fast approach for pattern recognition based on minimum bounding rectangle. For the third module, the recognition is made by using a matching directly between the SURF descriptors of the detected traffic sign and the traffic signs in the database, in this module we apply a filtering interest points detected and we keep only the points that are inside the pictogram's sign. The evaluation of the proposed approach gives good results compared to some powerful techniques. As a result, with the proposed system we have obtained a high performance with 95.65% sign detection, 97.72% traffic sign identification and 89.59% traffic sign recognition rate in an average time less than 80 ms/image.