Need help with ANPR Solution!

,
1

Hello Team,

I am working on developing an ANPR(Automatic Number-plate Recognition system) solution. The Yolo V8 model is giving me an accuracy of 97% in detecting the number-plate. But the OCR models (EasyOCR, PaddleOCR) are not giving me a good accuracy in recognizing the number-plate. I am planning to custom train the Tessearct OCR 5 for number-plates, but its Cumbersome. Can someone help me with some good ideas and approach for OCR??

2

Maybe using Pytesseract will be better than the OCR-based solutions (Easy, Paddle, etc). At least it will be faster.

3

it’s all OCR. all of it can be trained.

tesseract is generally lagging behind the state of the art and the competition (easyocr, paddle, …). at least, I judge that by how successful people are with anything non-trivial they need to read.