We have lifeguards. Less sexy (well, in a strictly technology-enthusiastic context), but proven in use.

The comments in this thread would suggest otherwise though.

Life guards are a lot like teachers, the ratio of eyeballs to swimmers matters. A 100 to 1 ratio is impossible for any one person to track.

Technology could help "flag" swimmers for human inspection (lifeguards, in this case). But any "timeout-based" approach is a no-go. I used to be able to go underwater for about 3 minutes (less depending on exertion), while I doubt my mom would be able to hold her breath for more than 10 seconds before starting to drown.

Something like those battery powered Finger Pulse Oximeters that doctors use comes to mind; they use light to measure oxygen levels and heart rate, and they're super cheap at $10-$30.

This is like trying to do services monitoring by just looking for service time-outs - seems intuitive but lots of false positives and by the time you observe it the disaster is in motion. A watchdog timer or deadman's switch is used in a lot of services as an alternative but would be a UX problem since it would require a form of active participation by the users. On the other hand, Disneyland had some success with tracking devices of sorts by branding them right as something beneficial to parents as well as children, so this barrier may be easy to overcome.

I'm also curious if computer vision algorithms could help with anomaly detection scenarios like this when the automatic response is so universal and there might be a large body of videos and photos of such events to help train machine learning algorithms upon.

Lots of pools have such systems, e.g. https://en.m.wikipedia.org/wiki/Poseidon_drowning_detection_...

It's a straightforward but very meaningful application of computer vision.