Pengoptimalan Deteksi Plat Nomor Kendaraan di Lingkungan Cahaya Rendah dengan Restormer dan YOLO

Abstract

The method used in this study combines Restormer and You Only Look Once (YOLO) to address the problem of vehicle license plate detection under low light conditions. Restormer functions as an image enhancement stage to improve illumination without losing important details, while YOLO is utilized to detect license plate areas quickly and accurately. The dataset consists of 1,000 vehicle images captured at different times of the day morning, afternoon, evening, and night. Which were processed through data augmentation and divided into training, validation, and testing sets to ensure balanced learning. The experimental results show that applying Restormer as a preprocessing stage significantly improves the YOLO model’s performance. The precision increased from 0.733 to 0.782, recall from 0.673 to 0.719, mAP@0.5 from 0.698 to 0.754, and mAP@0.5–0.95 from 0.434 to 0.487, with an F1-score of 74.9%. These results demonstrate that the integration of Restormer and YOLO effectively enhances vehicle license plate detection in low-light environments. The developed system is expected to serve as a foundation for future implementations in traffic monitoring and automatic parking systems that require high accuracy under various lighting conditions.