Believe it or not, cameras are exactly how the existing system works. I can do it now with OpenCV, but I've thought about looking into doing the same thing with mmwave radar instead. I think the radar route is more data-rich, but obviously more complicated and expensive. In the meantime, openCV + a webcam pointed at the pins, then we do some really basic object detection to detect pinfall.

While reading all about cameras and other sensor I would try to use distance-driven sensors every where. Standing pin vs laying one have enough height difference to not mislead even simple distance echo-based sensor. Other techs I see more reliable are LIDARs, because they don't require lighting. Pure camera should have enough light to see the pins (or you have to do some IR based markings and IR camera) or you can't make the space darker for events etc.