// the find
liuruoze/EasyPR
(CGCSTCD'2017) An easy, flexible, and accurate plate recognition project for Chinese licenses in unconstrained situations. CGCSTCD = China Graduate Contest on Smart-city Technology and Creative Design
EasyPR is a C++ library for recognizing Chinese license plates in real-world photos, built on OpenCV with SVM for plate detection and ANN for character recognition. It targets developers who need to add plate recognition to Chinese-market applications and don't want to build the CV pipeline from scratch. Last meaningful work was done around 2017; this is effectively a historical artifact.
The pipeline is cleanly decomposed — plate locate, plate judge, chars segment, chars identify are separate stages you can swap or debug independently. Pre-trained models are included so you can run inference without sourcing your own training data. The test dataset is reasonably sized (250+ real-world plate images across many provinces) with XML ground truth, which makes benchmarking straightforward. Multiple detection strategies (CMSER, Sobel, color) can be combined at runtime, which is useful for handling different lighting conditions.
The ANN/SVM stack is frozen in 2017 and has never been replaced with a CNN, which the README itself acknowledges as outstanding work. 80% accuracy on clean images sounds acceptable until you realize that means one in five plates fails in ideal conditions — unconstrained real-world performance will be worse. No support for new-energy vehicle plates (green background, different format), which now make up a significant share of Chinese traffic. The OpenCV 3.x version dependency warts are not cleanly handled — there are compile-time `#define` switches you have to find and flip manually rather than the build system detecting the version.