The Korean License Plate Detector is a real-time image processing pipeline based on ONNX Runtime that leverages an object detection model to locate license plates and recognize individual characters. Unlike traditional character recognition methods, it uses an approach that detects each character as an independent object, achieving a high recognition rate even in challenging conditions like blurry images, tilted angles, or partial obstructions. It consists of a three-stage pipeline that sequentially applies three specialized models to detect the license plate area, perform perspective correction, and finally recognize the characters.

The GUI based on PySide6 supports batch processing by directory and provides a flexible interface for manual input in case of detection failure. The processing results are automatically saved as Excel files, and single executable file distribution using PyInstaller makes it readily usable in the field.