Abstract
Traffic signs occupy a small proportion in the image, which makes feature extraction difficult. This study focuses on small prohibitory signs. First, we introduce super resolution (SR), which can improve the quality of the image, to solve the recognition problem of small prohibitory signs. Second, we propose a new model that can detect and classify various prohibitory signs compared with a traditional joint model. The proposed model utilizes the color and shape of prohibitory signs in generating proposals and filters the negative samples. The retained proposals are divided into small, medium, and large objects based on the size of prohibitory signs, and the small and medium objects are reconstructed through SR. Finally, the proposals are classified by using support vector machine (SVM) algorithm. Experiments on Tsinghua-Tencent 100K and German Traffic Sign Recognition Benchmark (GTSRB) datasets demonstrate that the proposed method is feasible. The proposed method achieves recognition accuracies of 78% and 67% on the small objects of Tsinghua- Tencent 100K and GTSRB datasets, respectively, which exceed the recognition accuracy without SR.