Fog-resilient Bangla car plate recognition using dark channel prior and YOLO

Abstract

Despite advancements in Automatic License Plate Detection (ALPD) methods, the majority of them fail to address the diverse image challenges faced in real-world driving scenarios. These challenges encompass issues like low image quality, contrast issues, etc. Factors such as license plate background, horizontal tilt, and adverse weather conditions like rain or fog further impede LP detection and recognition. This research focuses on the localization and recognition of Bangla vehicle plates in foggy conditions through the application of the Dark Channel Prior (DCP) fog-dehazing technique. The selection of Bangla as the target language is motivated by its status as a low-resource language with high digital text complexity, resulting in limited available resources. The proposed method comprises three main phases. The DCP dehazing algorithm reduces fog in input images initially. Then, the YOLOv8 object detection model is used to identify Bangla license plates from dehazed images, followed by OCR for text recognition. This study leverages DCP, YOLOv8, and OCR technologies to enhance the identification of Bangla vehicle plates under hazardous conditions, thereby contributing to the improvement of transportation safety, law enforcement, traffic management, and taxation processes.