Ask HN: What would be the best strategy to read game data from a video stream?

I want to extract game data from CSGO recorded matches programmatically. More specifically, I’d like to get each player’s position on the map during the game at any given time with perfect precision (0, 1s). I’m wondering what would be the most efficient way of pursuing this. I thought of learning and applying AI-powered image classification to achieve this, but given that the UI doesn’t change much from match to match, I wonder if there wouldn’t be a more straightforward path where I write a custom algorithm to do it. Would you have some insights? Maybe there’s similar software somewhere? Thanks!

* here is an example of a video: https://youtu.be/_iofTrXtPVs