Improving ALPR Stability with GAN-Based Super-Resolution for Iranian License Plates in Challenging Weather Conditions

Vehicle license plate recognition systems have garnered significant attention in the field of computer vision due to their crucial role in intelligent transportation systems. Despite advancements in hardware, including high-quality cameras, these systems often struggle to maintain performance under adverse weather conditions such as fog, dust, rain, and blurriness. This research addresses this challenge by leveraging Generative Adversarial Networks (GANs) as a generative model to learn the distribution of various anomalies in images and reconstruct them to mitigate their impact. This approach enhances the resilience of license plate recognition systems, enabling them to function effectively even in challenging weather conditions. The study also focuses on learning the distribution of real-world anomalies, a key motivation for this research. For character recognition on license plates, the YOLO network was employed, achieving an average accuracy of 97.4% with an inference time of 15.8 milliseconds. To evaluate the effectiveness of the GAN model, images were intentionally contaminated with three types of noise: fog and dust, blurriness, and rain, with noise levels ranging from 10% to 70%. Prior to applying the GAN-based reconstruction model, the character recognition system achieved average accuracies of 71%, 59%, and 87%, respectively, for these noise types. Post-reconstruction using the GAN model, these accuracies significantly improved to 94%, 85%, and 90%, respectively.