Comparing YOLOv5 and YOLOv8 Performance in Vehicle License Plate Detection

The number of mobile vehicles on the roads in Indonesia is increasing every year. Therefore, it is essential to verify the identities of these vehicles for a variety of reasons, including locating stolen vehicles, enforcing traffic laws, managing car parks, and collecting tolls. Nevertheless, inspecting these vast numbers of vehicles manually is a challenging task. Motor vehicle number plate detection and recognition play a crucial role in intelligent transport systems. Generally, the detection and recognition of number plates on motor vehicles entail three main stages. Machine learning-based object detection, which encompasses a range of algorithms that can automatically identify and locate objects in images or videos, is the first stage. These models leverage multiple layers of processing units to extract intricate features from input data, thereby enhancing overall efficiency for object detection purposes. The YOLO algorithm is a popular object detection algorithm that can detect objects from images or videos in real-time using custom dataset. In this study, we directly compared YOLOv5 and YOLOv8 models which underwent equal training epochs, achieved stability, and utilized hyperparameters with an image size 640, 100 epochs, val 200, and batch 16. The YOLOv8 gets the best performance with almost 97.5% mAP and 69.4% mAP50-95. Keywords: Plate Number, YOLOv5, YOLOv8, Object Detection, Custom Dataset