Transfer Learning with ResNet50 for Robust Traffic Sign Recognition: Addressing Challenges in Lighting, Occlusion, and Class Imbalance

The efficacy of deep learning models, specifically Convolutional Neural Networks (CNNs), in accurately recog- nizing and clas- sifying a diverse range of traffic signs. The research utilizes a meticulously curated dataset obtained from Kaggle, comprising annotated images labeled with corresponding ClassIds representing various traffic sign categories. The moti- vation behind this work stems from the critical need for robust and reliable traffic sign recognition systems, which play a vital role in enhancing safety, efficiency, and enabling autonomy in transportation systems. Through systematic experimentation and rigorous validation, the study assesses the performance of CNN models across various evaluation metrics, shedding light on the effectiveness of these deep learning architectures in accurately discerning nuanced visual cues present in traffic sign imagery. Furthermore, the research explores the challenges posed by factors such as variations in lighting conditions, occlusions, and class imbalances, proposing potential mitigation strategies. The proposed methodology utlized transfer learning model ResNet50 which achieved accuracy of 92