Enhancing Traffic Sign Recognition: A Deep Learning Approach for Occluded Environments

Atharva Yeola, Chandranath Adak, Soumi Chattopadhyay, Sukalpa Chanda

Abstract: In the modern era, technological advancements have surged, particularly in autonomous driving systems and advanced driver-assistance systems, where accurate traffic sign recognition is essential for safe and efficient navigation. However, detecting and classifying traffic signs accurately becomes challenging in real-world conditions due to occlusions caused by environmental factors, adverse weather, vandalism, and other visual obstructions. This paper presents a study into the issue of occluded traffic signs. Our study begins by assembling a diverse dataset of occluded traffic signs and then engages a transformer networkbased deep architecture for traffic sign recognition. To assess the effectiveness of our approach, extensive experiments were conducted on a curated dataset, benchmarked against several contemporary methods. The results demonstrated encouraging performance and showed robustness in handling occluded traffic signs.