Starling face recognition via Nest

9to5Mac: Starling Home Hub 6.2 adds Nest two-step verification, face recognition, and more.

That's pretty cool, especially the face recognition piece. I like.

Cool concept but unless the camera does depth recognition it’s scary to think someone with a photo of me could hold it up and unlock my door... most doorbell cams likely don’t have the right sensors to detect it’s a true 3D object.

1 Like

For sure. But for fun announcements it would be ok I think.

*** SUBMARINE DIVE SIREN ***
WARNING WARNING WARNING
MOTHER-IN-LAW INBOUND ON DRIVEWAY
*** SUBMARINE DIVE SIREN ***

:crazy_face:

4 Likes

Hah, yeah some things like that would be fine. I can see some cool uses. The article just happened to use the example I said of the Hello unlocking the door as you arrive home. That's a scary implementation to me.

2 Likes

Agreed! And even more scary if it let the inlaws in without any warning :smile:

Just thinking out loud here but shouldn't it be possible to create something on a rpi that intercepts the data going to homebridge and send it straight to Hubitat instead? Way over my head but it just seems that if the homebridge app can capture the data, why can't someone much smarter than I, come up with something to send it to our Hub. :thinking: :thinking: :grin:

1 Like

There are loads of examples on the Web of Raspberry Pi's running face recognition. So it's definitely feasible. But how robust and hence secure will be the question I think. But for a fun project it should be feasible....

Another thought (potentially at least for the future), I pinged the author of tinyCam the other day. It's an awesome app for web/ip cam management. It covers a huge range of cameras. He has motion detection on his app and this can send a get to an endpoint. So I've rigged this up on an Android tablet, with the built-in tinyCam webserver, to my new HE dashboard to show the real time video and then flick a switch in HE when motion is captured. But even better than that, his app has built-in recognition for vehicle, person, pet and face detection. I've asked him if he could set up separate endpoints for those triggers too. He said he would think about it :slight_smile: So this could also be a possible future option without resorting to a separate Pi and Python etc. I told him what we really need is number plate recognition and face recognition against a library of images using machine learning built into his app lol :smile: If my Nest Display can do it I'm sure it ain't that hard :stuck_out_tongue:

https://www.pyimagesearch.com/pyimagesearch-gurus/

Online training from one of the guys referenced above, covering number plate, face, hand gesture recognition/tracking. One day when I get some spare time...

This Android ​app is actually quite amazing...if you can conquer the appalling UI (which we should all be quite seasoned at on here right?) :smile:

First add a person (or 2+ideally) in training mode (20 photos are taken, and you can add more later if you like because they are just stored in the file system). Then select Recognition Training. Then finally detect the person using Recognition View.

All it needs is a Maker API endpoint get per person and we have a working solution! Any keen Android​App developers willing to add that to the open source? :thinking:

Hmmm, @Angus_M you seem to be fixated on just facial recognition... I was thinking in a much broader sense. The device you initially posted captures everything Google/Nest has to offer. To have this data available in HE would be a game changer.

Oh well, maybe someday!

1 Like

Ha ha ha no not really. It's just that the original post was referring to face recognition. As I mentioned above I'm working with the developer of tinyCam to get additional webhook's into his system for vehicle, pet, person, and face detection (he already has motion). Once this is done we will have a method to do recognition of all those across the vast array of web/ip cameras that he supports!!!