Face recognition / Dlib & RPI3

Hi all,

i would be interested in implementing face recognition with Dlib or similar in my home automation system.
it looks pretty easy to do but reading here and there it seems an RPI3 won’t be able to cope with the computational load.

can someone of you guys confirm this?
something i would like to do, for example, is trigger automation based on face identification which means, for my use case, i need a response time of about 2-3 seconds.

You can easily do this, please refer to my github for complete

Raspberry Pi 3 configuration and setup. (https://github.com/ammar-khan/raspberry-pi-3-model-b-plus)
Also you will find many face detection code in my github:
https://github.com/ammar-khan?tab=repositories

Feel free to contact me

Thanks

ehy thanks for that really!

I will have a look at it!

what kind of performance should I expect on an RPI3?

Hi,
What type of performance you are looking for?
May I ask what you wanna do using Raspberry Pi?
Cheers
Ammar

I would like to use it for face recognition of my doorbell camera to make such that:

  • if you ring the doorbell
  • if your are flagged as “home” by an ibeacon

the above 2 for extra safety

  • if the SW recognize your face

the door opens.

So I would like face recognition to happen within 5s say: input should be a single image and not a camera stream as my doorbell save a static image of the “last ring”.

Sorry for late reply.

If you want to do just one face recognition. You can do it with Raspberry Pi 3 Model B+, why I prefer this model because it has wifi and bluetooth built in. You can easily achieve max 2 frames per second with deep network which is more accurate.

Let me know if u need any help.

Can you please tell me what type if door bell you have? Does it has IP address? How you can get image from door bell? Also if u have written any code can you share it on github so I can suggest you possible tuning and optimization for Raspberry Pi

hi, i have doorbird 201.

this has an IP address and is compatible with HA.
it streams 3 “cameras”:

  • live
  • last ring
  • last motion

also it triggers an event to HA once you ring the bell.
So you could set up a script/automation which is triggered by doorbell ring and run face recognition on “last ring” stream which is, actually, a static image.

Except these considerations I had no time to set up anything so I can’t help further