Enhancing road sign recognition using Convolutional Neural Networks
Yuning Zhuang · Applied and Computational Engineering · 2024
The significance of road sign recognition technology has grown in tandem with the development of autonomous driving systems. However, traditional methods have encountered limitations when faced with intricate road scenarios. In contrast, CNNs have the capacity to autonomously glean complex features from image data, leading to a substantial enhancement in the stability and accuracy of road sign recognition. In this research, we employ a road sign dataset obtained from Kaggle and apply image preprocessing techniques, coupled with data augmentation methods. Furthermore, we employ transfer learning by utilizing ResNet34 to construct a robust road sign recognition model. The experimental outcomes vividly demonstrate the efficacy of this model in acquiring intricate features of road signs, resulting in the augmented accuracy and robustness in recognition. Nevertheless, we do observe certain limitations, including recognition instability and shortcomings in generalization. Nonetheless, on the whole, CNNs exhibit significant potential for application within the realm of road sign recognition.