Road Sign Recognition Using Convolutional Neural Networks

Road sign recognition is critical for autonomous driving and advanced driver assistance systems, ensuring road safety and efficient traffic flow. This paper presents a study on developing an accurate and robust road sign recognition system using convolutional neural networks (CNNs). The study explores various CNN architectures, training techniques, and data preprocessing methods to optimise performance. A detailed analysis of the Traffic Signs Preprocessed dataset is conducted, and a series of nine CNN models with different filter sizes are trained and evaluated. The results demonstrate the effectiveness of CNNs in extracting relevant features from road sign images and accurately classifying them into standard categories. The study also investigates the impact of filter size on model accuracy, providing valuable insights into the trade-offs between complexity and performance. Additionally, the paper discusses implementing a software application that integrates the trained CNN model for real-time road sign recognition from images and videos. The application's graphical user interface allows users to upload data and visualise the detected and classified road signs, showcasing the practical applicability of the developed system.