Road sign classification system using cascade convolutional neural network

We proposed a road sign classification system using C-CNN (cascade convolutional neural network) classifier. The cascade configuration is designed so that the classifier can easily converge with the data. Our system consists of six stages of Network in Network (NiN) architecture based CNN classifier. The data augmentation method is used to enrich the training and testing dataset which also tests the robustness of our system,. Our Japan road sign dataset consists of ten classes with 7,500 examples for each class. Each image cropped from real street images is taken by the camera attached to the top of the car. From the experiments, our system is more efficient compared with bag-of-features method. The execution time of our system is less than 20 ms using appropriate hardware configuration, which is suitable for real-time application approaches like an autonomous car or driver assistance system.