Learning Subjects For Home Security

I was thinking of a set of features that I wish existed that would make installations much easier. I was reading the "Worlds Smallest PIR" article when an idea hit me. Now this my not be such a new concept to many, but it was something that I "dreamed up" myself as I read that article.

What if your analytics could "learn" subjects, either by facial recognition, or an RFID chip/sensor combo, or any other way of identifying subjects in a scene? You could then trigger events, such as motion recording or alarm outputs based on the individual subject.

For instance, at home, you could have the system "learn" the homeowners and it would ignore their activity in the home. However, if a stranger (aka intruder) entered the home, it would activate motion / alarm output / etc. You could have a grace period to allow a guest into the home, or set by time, or allow the system to be set to monitor only, no alarm. There could be many different types of states of awareness you could program.

1) Full alert - motion and alarm instantly when an unknown person enters. This would be useful at night or when you are away from home.

2) Delayed alert - motion and alarm with a set time delay for the homeowner to disarm if needed. This would be nice if you had an unexpected visitor during the daytime. You could have it give that visitor a predetermined amount of stay time, ex 2 hours, before it would alert you again of a unknown person.

3) Motion only - motion record only, no alarm. This would be useful when having guests over, or maybe a repairman. It would not trigger the alarm, but would still record their activities.

4) Stand down - basically the system would not record or monitor activity at all.

Now all of this would require the ability to "learn" and identify subjects. With the tracking ability that we saw in the nursing home study, I don't think it is too far out of reality to expect this type of ability in the near future. Processing power is there to run software. You would just need to find a way to identify subject reliably.

What do you think? Let's see where we can take this discussion and see how viable it is.


Panasonic does facial recognition on their top of the line Viera HDTVs couple with voice command. You can customize a personal home screen with the TV apps you use (netflix, internet, pandora, etc.) Pretty fun stuff.

Disclaimer: My paystubs have a Panasonic logo on them....

Good point about the Xbox One! I knew about that, but forgot about the recognition features. If they can do it with a 1080p webcam, we should be able to do it with well placed cams at each entrance, then tag the subject, and follow their movements through the house, just like in the nursing home study.

I like it -- a brilliant and original idea that's sure to be worth a ton of money.

... oh, wait ... Try Googling Xbox One ... Microsoft thinks they've got this working for viewer-specific content. How to make it extensible, anyone?

Maybe set top boxes have a better tolerance for the odd identification error. For example, how bad can it be if I occasional get exposed to my Wife's set top box home page?

I'll say this has been the "holy grail" for home automation for many years now. I'd saw that we're still a long way away from this being truly feasible (ie: it's reliable, affordable and easy to maintain) if it's dependant on any sort of facial recognition. It's slightly more practical with RFID or a similar thing, but the WAF on solutions like that is pretty low.