Deep Residual Learning and U-Net Segmentation for Automated Traffic Sign Detection and Classification

The manual inventory management of the traffic signs is proposed to be replaced by an automated traffic sign detection and recognition framework as per the study. The system comprises U-Net for the precise segmentation of traffic signs from complicated road surroundings and employs CNN and ResNet152V2 models for multi-class classification. To ensure robust training and generalization under different conditions, a dataset of ,000 augmented images in 52 traffic sign categories was used. The CNN model resulted in a validation accuracy of 82.1%, whereas the ResNet152V2 network, which was made deeper with the residual connections for feature extraction, achieved a higher 98.3% accuracy. The main reason for the better performance of ResNet152V2 over traditional CNN architectures in the experiments is the improvement of stability during convergence and the reduction of overfitting. Deep residual networks are one of the most effective solutions for the traffic sign recognition problem in real-time scenarios, as this work demonstrates. Furthermore, it provides a basis for their use in Advanced Driver-Assistance Systems (ADAS) and autonomous vehicle applications, thus, increasing safety and the efficient management of traffic.