Traffic sign recognition plays a critical role in autonomous driving and advanced driver-assistance systems (ADAS). This project leverages the power of transfer learning by fine-tuning a pre-trained AlexNet model on the German Traffic Sign Recognition Benchmark (GTSRB) dataset. A custom-built GUI was also developed for real-time classification of traffic signs. The trained model achieved an impressive 95% accuracy with minimal training epochs. Evaluation metrics included accuracy, loss, precision, and recall. The system showcases how deep learning and intuitive interfaces can be combined for practical real-world applications.
The AlexNet-based model provides accurate and fast classification of traffic signs, making it suitable for integration into intelligent transportation systems. The GUI enhances usability, demonstrating how AI models can be deployed in accessible formats. Future work includes experimenting with deeper networks like ResNet and integrating real-time image capture for live classification.
The model is trained only on the GTSRB dataset and may not generalize well to signs from other countries or under poor lighting conditions. Further data augmentation and domain adaptation are necessary for broader applicability.
Plans include improving GUI design, expanding dataset diversity with international signs, and deploying the system in embedded hardware for real-time vehicle use. Comparative studies using other architectures like ResNet and EfficientNet are also underway.