A Robust Network for Embedded Traffic Sign Recognition

Traffic sign recognition systems are vital in real-world applications such as auto-driving and safety and driver assistance. While deep neural networks have achieved high accuracy in classifying traffic signs in recent years, there is always the discussion of the high computations of these networks and their many teachable parameters. A significant challenge is to design a compact deep neural network for the application of traffic sign recognition. This paper proposes a network that uses residual blocks in the network to obtain a top-1 accuracy of 99.51 for the German traffic sign recognition benchmark. The number of parameters is ~430,000, which is ~32x fewer than the state-of-the-art. Experiments have been performed to show the network's resistance to destructive factors and its comprehensiveness in the application of traffic sign recognition. These tests show that it is a comprehensive and robust network for the recognition of traffic signs.