Deep Learning-based Recognition System for Libyan License Plates

Automatic license plate recognition system plays a crucial role in a variety of practical applications, particularly those pertaining to security and traffic management. Its primary function lies in the extraction and identification of numerical plate data from videos or images of the designated automobile. It is important to note that vehicle license plates differ across countries, which affects how well any particular system or technique works depending on the kind of plate. Within the context of this study, we introduce an automated recognition system specifically designed for Libyan automobile registration numbers. The primary barrier encountered in this endeavor is our commitment to utilizing authentic vehicle plate images from Libya, the majority of which exhibit suboptimal conditions due to inadequate vehicle maintenance. Two deep learning techniques, namely YOLOv8 and Roboflow object detection, were compared in this study to determine the optimal method for locating authorized license plates. The process of plate character recognition utilized the EasyOCR method. To evaluate the efficacy of the suggested system, an assessment was conducted on a collection of 2400 vehicle images encompassing diverse lighting circumstances and backgrounds. The Roboflow object detection technique exhibited the highest performance in plate detection within the utilized dataset, achieving precision and recall values of 99% and 100%, respectively. Conversely, the accuracy performance of the EasyOCR method in recognizing plate characters reached a mere 64.6%.