Optimize Python Script Performance

I have an existing Python YOLO/EasyOCR script that works but produces a lot of garbage, as well as the required correct reading.

I want someone to provide an improvement that keeps the most 'likely' value, and drops the bad readings.

Since the values are coming from video frames, I expect that the correct value is read many more times than random false ones, so the task is to identify the OCR values that occur most frequently during the video sequence and only store those. In addition, the values need to be associated with the relevant roi boxes that they come from. Alternatively - if you have a better way of identifying the 'real' value, please discuss your approach.

It would be a bonus if you can add the 'best' value to the video on the relevant roi box, but not essential.

What I expect from you: refactor or rewrite the relevant functions, leverage efficient data structures or operations, and leave clear inline comments so it’s easy for me to maintain. Demonstrate the improvement with supplied video footage.