REAL-TIME AUTOMATIC VEHICLE REGISTRATION PLATE RECOGNITION FOR PAKISTANI TRAFFIC USING YOLOV8S AND OCR

Abstract

Automatic license plate recognition, or ALPR, plays a key role in intelligent transport and traffic law enforcement. Most conventional ALPR techniques struggle when lighting is poor, plates are tilted, or when dealing with the varied formats used on Pakistani vehicles. Given these challenges, we developed an end-to-end framework combining YOLOv8s for plate detection with OCR for character recognition. We trained and tested the model on a Roboflow dataset of 3,000 Pakistani vehicle images collected under different lighting, occlusion, and angle conditions. On the test set, the system reached 94% mAP@0.5, 94% precision, and 94% recall, while running at 28 FPS on a GPU, which makes real-time use feasible. The results show the model handles low-contrast scenes well. However, it still faces challenges with badly damaged plates, extreme angles, and very fast-moving vehicles. Overall, YOLOv8s proves to be an efficient and accurate choice for traffic monitoring and automated tolling in complex, real-world traffic.