I worked on that.
It sends audio and/or video fingerprints (not frames, for privacy and bandwidth reasons), which are matched against a fingerprint database. Whatever people see on TV is usually 10 to 60 seconds behind the real live stream at the broadcaster (which is where the reference fingerprinting happens). GeoIP data can be used to roughly deduce where the TV is located, in order to better filter out false positives out of multiple matches (e.g. in the US where lots of programming on east/west side is just shifted by ~3 hours due to time difference)."
" xcgfhgjbjhb on Sept 19, 2019 [–]
I worked on that. It sends audio and/or video fingerprints (not frames, for privacy and bandwidth reasons), which are matched against a fingerprint database. Whatever people see on TV is usually 10 to 60 seconds behind the real live stream at the broadcaster (which is where the reference fingerprinting happens). GeoIP data can be used to roughly deduce where the TV is located, in order to better filter out false positives out of multiple matches (e.g. in the US where lots of programming on east/west side is just shifted by ~3 hours due to time difference)."
From this thread - if you find it interesting https://news.ycombinator.com/item?id=21002745