Semantic segmentation-based traffic sign detection and recognition using deep learning techniques

We present a method for detecting and classifying traffic signs based on two deep neural network architectures. A Fully Convolutional Network (FCN) - based semantic segmentation model is modified to extract traffic sign regions of interest. These regions are further passed to a Convolutional Neural Network (CNN) for traffic sign classification. We propose a novel CNN architecture for the classification step. In evaluating our approach, we contrast the efficiency and the robustness of the deep learning image segmentation approach with classical image processing filters traditionally applied for traffic sign detection. We also show the effectiveness of our CNN-based recognition method by integrating it in our system.