A New Approach for Traffic-Sign Recognition using Sparse Representation over Dictionary of Local Descriptors

This paper is meant as the object recognition work done directly on images obtained from the autonomous car’s camera. Normally, before the recognition stage, images need to process with the purpose of locating the candidate regions. This paper does not concentrate on the location stage, but focus on contributing a new descriptor for object recognition. Particularly, objects such as traffic-sign are recognized based on the sparse representation of object’s descriptor over learned dictionary. The main idea is training the dictionary of local descriptors, or building the basis of a local descriptor space. Then, using the representation of description of object over this basis, we define a object’s descriptor. We approve the robustness of the proposed descriptor under the linear trans- formation such as rotation or scale etc. Moreover, the stage of building the basis of descriptor space is done offline, thus computing time is quite reliably for the autonomous car. The experiments done on a database including traffic-sign images show that the proposed method is promising and competitive related to the state-of-the-art methods.