Comparative Experiment and Performance Evaluation of Three Different Models of Traffic Sign Recognition System

Traffic sign recognition is one of the key tasks of intelligent transportation system, which is of great significance to improve traffic safety and efficiency. As deep learning technology continues to advance, more and more studies begin using deep learning models to improve the accuracy of traffic sign recognition. Aiming at the traffic sign recognition task, this essay uses the GTSRB dataset to construct three deep learning models, CNN, MobileNet and ResNet. The experimental results show that in traffic sign recognition tasks, ResNet achieves the highest accuracy of 0.9970, leveraging residual learning for superior feature extraction, making it ideal for high-precision requirements. CNN follows with an accuracy of 0.9827, showing a good balance between performance and simplicity, while MobileNet, though less accurate at 0.7540, excels in efficiency and lightweight design, making it suitable for resource-constrained environments. These methods are designed to address current limitations and further advance the development of traffic sign recognition technology.