Detection and Identification of Vehicle License Plates in Indonesia Transportation System Based on Deep Learning Using YOLOv11 and Easyocr

The detection and identification of vehicle license plates in Indonesia still face significant challenges due to unstable environmental conditions during image capture, such as extreme lighting, varying angles of capture, physical damage to the plates, and diversity in design and font types. These conditions degrade the accuracy of existing recognition systems, especially if the model is not trained to handle such variability. In addition, the public's low understanding of license plate structure also hinders the optimal use of this information. This study aims to develop an accurate, adaptive license plate recognition system for real-world conditions that can interpret license plate information in real time. The model was created using the YOLOv11 algorithm for fast, high-precision plate detection, and EasyOCR for plate character identification. The dataset consisted of 709 images of two-wheeled (motorcycle) and four-wheeled (car) vehicle plates, collected from public datasets, the researchers' surroundings, and the campus area. Most of the data was collected through direct photography with cell phone cameras, reflecting real-world field conditions. The test results show that the YOLOv11 model has excellent detection performance, with mAP@50 of 94.2%, precision of 97.7%, and recall of 86.7%, while the EasyOCR method achieved a character recognition accuracy of 91.0%. The main innovation of this research is the application of a license plate recognition system to support intelligent transportation systems in campus environments, particularly for parking system implementation.