Optimized YOLOv8 for Automatic License Plate Recognition on Resource Constrained Devices

This paper presents an optimized Automatic License Plate Recognition (ALPR) system designed for resource-constrained devices, leveraging YOLOv8 for real-time object detection and Optical Character Recognition (OCR) to extract license plate information under challenging conditions such as low-light, motion blur, and occlusions. Unlike traditional ALPR systems that rely on high computational resources, our approach balances detection accuracy, processing speed, and efficiency. The system is evaluated on three benchmark datasets: the Chinese City Parking Dataset (CCPD) with 250,000 images under diverse conditions, the UFPR-ALPR Dataset (Universidade Federal do Paraná, Brazil) containing 4,500 real-world traffic images, and the RodoSol-ALPR Dataset with 20,000 highway surveillance images for high-speed license plate recognition. Among various YOLOv8 variants tested, YOLOv8-s achieved the best performance, with a mean Average Precision (mAP) of 99.3% while sustaining over 30 Frames Per Second (FPS), making it suitable for real-time ALPR applications. Furthermore, image sharpening and contour segmentation techniques improved OCR recognition accuracy by 5.1% under low-light conditions, improving robustness. Comparative analysis against state-of-the-art OCR-based ALPR methods (EasyOCR, FastOCR, and CR-NET) demonstrated that our approach surpasses existing models in both recognition rate and computational efficiency. These findings establish YOLOv8 as a highly effective and deployable solution for intelligent transportation, surveillance, and law enforcement applications requiring real-time license plate recognition with minimal computational overhead.