Need help with OCR accuracy

[Ryan Bijoy]

I am working on extracting scoreboard data from game images in real-time, with high accuracy and speed. The template and language (English) stay the same, but fonts may vary (ignore the one Chinese name).

I annotated the data in Roboflow, trained a YOLO model, and achieved 99% detection accuracy. For text extraction, I used Tesseract-OCR with some preprocessing, the results are good but it still misses some digits.

What can i do to increase the accuracy of the extraction?