Fast and Accurate Bangladeshi License Plate Identification using YOLO

Technological advancements have made an intelligent traffic management system crucial for society. One key part of this system is Automated License Plate Identification (ALPI). This research uses the Bangla LPDB-A dataset to carry out various CNN operations that contain diverse and challenging images of Bangladeshi license plates and presents a fast and effective transfer learning technique using the state-of-the-art CNN detection model YOLOv8 to identify Bangladeshi vehicle license plates. Dataset corrections in the preprocessing stage, custom augmentation applications, and background removal significantly boost detection accuracy, as demonstrated through comparative analysis against similar studies that do not apply dataset corrections and background removal, highlighting the superior performance of the proposed method. Using the transfer learning method on the YOLOv8 detection model, an impressive mAP(50) of 98.862%, mAP(50-95) of 77.46% and complete recognition accuracy of the license plate of 95.518% are achieved.