End-to-end-for-chinese-plate-recognition

Why we included this project

This project bundles the whole Chinese license plate recognition pipeline into one codebase: finding the plate in a scene, straightening its perspective, and reading the characters end to end. A U-Net segmentation model and OpenCV edge detection handle plate localization and rectification, then a CNN does multi-label character recognition on the corrected image. The author tuned it for the harder cases, like tilted camera angles, strong exposure, and dim lighting, where even some commercial recognizers fail. Everything is written in Keras on TensorFlow, so teams already on that stack can follow the training and inference flow without switching frameworks. It is a solid starting point for prototyping vehicle-related vision features, though the pinned Python 3.6 and TensorFlow 1.x environment means you will likely need to port it to a modern stack before production.