Abstract

An essential component of Autonomous Vehicular Technology (AVT) is the automatic classification of traffic signs. In order to keep autonomous vehicles safe and effective, traffic signs are essential. Consequently, it is critical to thoroughly evaluate the advantages and disadvantages of automated traffic sign classification (TSC) systems. The classification of traffic signs is a well-studied subject in the field of computer vision. The great majority of current methods work effectively on the traffic signs required for AVT. But, the existing classification algorithms failed to correctly classify the images which might vary negatively due to lighting, vehicle angle, vehicle speed variations, and other factors in real-time traffic. This is because the kind and severity of difficult conditions, such as high beams, rain, snow, fog, etc., are limited in the datasets of traffic signs that are now available. To handle these challenges, this paper proposes two approaches, one is the Climatic-Generative Adversarial Network (C-GAN) and the other is the X-Vision Transformer (X-ViT). The proposed C-GAN generates synthetic traffic sign images affected by challenging conditions such as rain, snow, high beam, fog, etc. The proposed X-ViT is used to correctly classify the real-time traffic sign images affected by various challenging conditions. The proposed models help to achieve state-of-the-art results on the benchmark datasets for TSC.

Authors

Joe Dhanith P R; Shravan Venkatraman; Abeshek A; Santhosh Malarvannan; Shriyans A; Jashwanth R