How to implement scraping not from the page, but from the video?

Question

Python

Good day to all! The essence of the question - there is a video that contains useful information (either words appear in a predefined part of the screen, or objects from a predefined set of objects). My goal is to write an application that will "see" and recognize this information (the text from the video will be transferred to plain text for processing, objects will be recognized and transferred again as text (name of the object) for processing). I ask for help in the direction for solving such a problem, at least literature (highly desirable, in Russian), especially, literature on neural networks is interesting. Maybe someone has come across something similar and there are implementation ideas - it's also interesting to listen, because my idea of ​​\u200b\u200bsolution is still rough and not optimal (and, possibly, unrealizable). Thanks for any help!