Enhancing Traffic Sign Recognition Via Transfer Learning: A Decision-Aid Framework with Real-Time Embedded Validation

This paper proposes a decision-aid framework for traffic sign recognition (TSR) based on transfer learning for real-time classification in autonomous vehicles and driver assistance systems. In this approach, a pre-trained convolutional neural network (CNN) model, fine-tuned with the GTSRB dataset, enables an efficient and accurate recognition rate for regulatory, warning, and priority traffic sign categories. To enhance processing speed and robustness under diverse environmental conditions, a lightweight transfer learning model is developed to optimize system resources. This framework is implemented and validated on Raspberry Pi and Nvidia Jetson Nano embedded platforms, achieving an average processing time of 120 ms per frame and an accuracy of 98.5%. These results demonstrate the effectiveness of the proposed model in TSR applications, aiming to improve road safety and driver assistance.