Authors: Atharva Rajesh Gosavi
Abstract: raffic sign recognition (TSR) is a critical component of autonomous driving systems, enabling vehicles to understand and respond to road regulations in real time. Traditional TSR approaches typically separate classification and localization tasks, resulting in increased computational cost and reduced robustness in complex driving environments. This paper proposes a multi-task deep learning framework that performs simultaneous traffic sign detection, classification, and attribute prediction using a shared feature-extraction backbone. The model leverages multi-task learning to exploit interrelated features across tasks, improving overall accuracy while reducing inference time—an essential requirement for self-driving applications. Extensive experiments conducted on benchmark datasets such as GTSRB and GTSDB demonstrate that the proposed approach outperforms single-task baselines, achieving higher precision in both recognition and localization. The results show that multi-task learning enhances generalization under challenging conditions, including occlusion, varying illumination, and high-speed motion. This work highlights the potential of unified deep learning architectures to deliver efficient and reliable traffic sign recognition for next-generation autonomous vehicles.
