Implementasi Metode YOLOv10 dan EasyOCR untuk Rekognisi Plat dan Identifikasi Sepeda Motor yang Bergerak

Abstract

The advancement of image recognition and computer vision technology has enabled the development of more intelligent and efficient traffic monitoring systems. One of the main issues in Indonesia is the limitation of the ETLE-INCAR system, which is not yet capable of automatically recognizing vehicle license plates, especially for motorcycles in motion. Under conditions of high speed or varying lighting, license plates are often difficult to capture accurately, even though such information is critical for identifying traffic violations. This research produces a system capable of detecting and recognizing motorcycle license plates in real-time using the YOLOv10 algorithm for object detection and EasyOCR for character recognition. The dataset used consists of 4,587 motorcycle license plate images from the Roboflow platform and 50 test images collected from real-world scenarios. The system was trained using the lightweight YOLOv10n model and evaluated using the Character Error Rate (CER) metric. The implementation results show that the system is able to recognize license plates accurately on moving objects and is deployed through a desktop-based application for easier real-time monitoring. After conducting the testing phase, this study achieved a YOLOv10 detection accuracy of 95% and a Character Error Rate (CER) of 17.2% for EasyOCR.