Traffic Sign Recognition Using Color and Spatial Transformer Network on GPU Embedded Development Board.

Traffic sign recognition is an integral part of any driver assistance system as it helps the driver in taking driving decisions by notifying about the traffic signs coming ahead. In this paper, a novel Architecture is proposed based on Convolutional Neural Network (CNN) for traffic sign classification. It incorporates Color Transformer Network and Spatial Transformer Network (STN) within CNN to make the system invariant to color and affine transformation invariant. The aim of this paper is to compare the performance of this novel architecture with the existing architectures in constrained road scenarios. The performance of the algorithm is compared for two well-known traffic sign classification dataset: German Traffic Sign dataset and Belgium Traffic Sign dataset. The paper also covers the deployment of the trained CNN model to Jetson Nano GPU embedded development platform. The performance of the model is also verified on Jetson Nano development Board.