E.g. in football/"soccer" the half time break isn't at a predetermined point in time, as the start of the game can be delayed and the referee can and usually does give some extra time to make up for unplanned breaks due to fouls etc. Same after regular time in the second half. And if it's a knockout game, then you may even have real overtime and even a penalty shootout to find a winner. Then you have unplanned/semi-planned breaks like players getting a break to drink when it's hot, or... heavy rain or... the ref getting hit in the head with a bottle. And people will use these breaks to do their "business" and make tea/coffee/use the microwave oven. And after the game, players may stay on the field to celebrate or cry about the loss, which I would guess a computervision "AI" would have trouble distinguishing from the actual game.
Just having some people watch the game in the control room (which they probably would do or at least want to do anyway) is still easier and more reliable than trying to train an AI to detect all that, in my humble opinion :D
Of course, automation like automatically detecting and scaling the grid by looking at the grid itself would and does help a lot; but that's different from automating watching the telly.
I have to admit that I was asking that as a loaded question as my thinking was much like yours, but I didn't feel like typing it out. A human is probably the most reliable "AI" for the job :D.
E.g. in football/"soccer" the half time break isn't at a predetermined point in time, as the start of the game can be delayed and the referee can and usually does give some extra time to make up for unplanned breaks due to fouls etc. Same after regular time in the second half. And if it's a knockout game, then you may even have real overtime and even a penalty shootout to find a winner. Then you have unplanned/semi-planned breaks like players getting a break to drink when it's hot, or... heavy rain or... the ref getting hit in the head with a bottle. And people will use these breaks to do their "business" and make tea/coffee/use the microwave oven. And after the game, players may stay on the field to celebrate or cry about the loss, which I would guess a computervision "AI" would have trouble distinguishing from the actual game.
Just having some people watch the game in the control room (which they probably would do or at least want to do anyway) is still easier and more reliable than trying to train an AI to detect all that, in my humble opinion :D
Of course, automation like automatically detecting and scaling the grid by looking at the grid itself would and does help a lot; but that's different from automating watching the telly.