The Korean license plate detector is a real-time image processing pipeline based on ONNX Runtime, utilizing an object detection model to locate license plates and recognize individual characters. Instead of the traditional character recognition method, this approach detects each character as an independent object, achieving high recognition rates even in challenging conditions such as blurry images, tilted angles, and 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.

With a GUI based on PySide6, it supports batch processing by directory and provides a flexible interface allowing users to manually input data in case of detection failure. The processing results are automatically saved in an Excel file, and the single executable file deployment using PyInstaller enables immediate use in real-world scenarios.