A Traffic Sign Recognition Algorithm for ADAS based on CNN for Complex Scenarios

The Traffic Sign Recognition (TSR) system is an essential component of Advanced Driver Assistance Systems (ADAS), which assists drivers in detecting and understanding traffic signs. However, recognizing traffic signs in complex situations, such as low or high lighting, adverse weather conditions, and background colour, remains a challenging task. While numerous researchers have achieved high accuracy rates, they have encountered difficulties in attaining a balance between recognition accuracy and processing speed. To address these problem, this paper introduces a TSR algorithm that uses Convolutional Neural Networks (CNN) with enhanced padding functions to increase recognition accuracy. The algorithm initially preprocesses the input image by performing contrast enhancement and normalization, and then trains the CNN model on GTSRB, a vast dataset of traffic signs to extract relevant features and classify them into respective categories. The padding function expands the input image dimensions, allowing the model to extract intricate features of background scenarios, increasing recognition precision. The proposed algorithm was evaluated on traffic sign images from real-world scenarios, including low or high illumination and adverse weather conditions, demonstrating its ability to achieve a balance between recognition accuracy and processing time, potentially enhancing the safety and efficiency of ADAS systems by providing timely and accurate information to drivers.