Challenging Weather for License Plate Detection: Impaired visibility

License plate detection is crucial for automatic toll collection, surveillance systems, and traffic management. This paper presents the findings of our investigation employing the YOLOv8 deep learning framework to detect license plates in challenging scenarios such as heavy haze, rain, snow, and sandstorms that severely limit camera operation and computer vision algorithms for traffic scene perception. Rain flow and ice reflections on roadways can cause enormous detection errors, which harm intelligent visual traffic systems. The major goal should be to achieve a high level of accuracy while maintaining real-time inference speed. To ascertain the dependability and robustness of the YOLOv8 system, a thorough assessment procedure was undertaken. Initially, a synthetic dataset was created to emulate challenging settings, such as the presence of haze and other meteorological phenomena, which are often encountered in real-life contexts. Thereafter, a simulation was conducted to evaluate the performance of the YOLOv8 model under poor weather conditions. Our method’s remarkable frame processing speed, reaching up to 70 frames per second on conventional hardware, renders it well-suited for real-time applications that need efficient and quick license plate identification. In comparison to established state-of-the-art methodologies, our approach demonstrates superior performance in terms of both accuracy and frame processing time, thereby representing a substantial advancement in the field of license plate detection. A variety of experiments have been undertaken on the CCPD dataset and its several subsets, including the Base, DB, Blur, FN, Rotate, Tilt, and Challenge datasets, which offer a diverse selection of license plate changes. The relevant CCPD subsets yield exceptional outcomes, with accuracies of 99.62%, 96.8%, 98.9%, 99.72%, 99.71%, 99.5%, and 99.1%, respectively.