Automated Vehicle Plate Number Recognition System using YOLOv4

This paper’s key contribution is an end-to-end automated license plate recognition pipeline — combining YOLOv4 with a CSPDarknet-53 backbone for vehicle/plate detection, classical image processing (thresholding, contour-based dilation) for character segmentation, and EasyOCR for alphanumeric recognition — deployed via a practical Streamlit web interface for real-world use (e.g., residential parking management). Notably, it presents a dataset-size ablation (Table I), showing that scaling from 347 to 737 images improved precision to 99% despite a slight mAP dip (89.2% → 83.26%), offering a transparent, honest account of the precision-recall trade-offs and current limitations (below-average training mAP/recall) rather than overstating performance.