Face and person detection with Deepstack - local and free!

Hi all
I just published a custom component for face and object (e.g. person) detection using Deepstack, which runs locally in Docker and costs you nothing. You will need a computer with a decent amount of RAM available, as Deepstack is using deep-learning models. Therefore don’t expect to run this on a pi, but a spare laptop should do. Any feedback on usage here, and bug reports on Github.
Thanks!

42 Likes

Cool.

Last line of your page (part of the faq), “tow” should be “two”.

2 Likes

This is pretty cool. Have you thought about individual recognition? Feed it sample pics of a person to learn and then report when that person is detected? This could be way better that geofencing.

3 Likes

Another excellent component @robmarkcole

I am running it alongside Tensorflow and should be able to compare the two.

2 Likes

Game-changing…thank you!

2 Likes

@BenDiss yes individual faces can be recognised but needs to be implemented

2 Likes

Sweet. I’ve got outdoor cameras at each entry point to the house. This could really perfect home/away sensing.

1 Like

I have been running the person detection part for 24 hours. Perfect and easy to setup.

Thanks

2 Likes

This looks amazing, thanks for the effort you put into this! I have some questions:

  • [edit] Is there a way to add the repo to hassio, so that it can be installed from within Home Assistant?
  • Will this work with Hassio, without proper SSH access?
  • Is the bare minimum spec-wise something like a NUC, purely for the RAM requirement?
  • Can the object detection be configured to detect car/car colour?

As others have said this is potentially an ideal presence detection solution, I guess I’ll be saving up for another NUC :slight_smile:

2 Likes

First post has github link.

@nickrout

First post has github link.

I wouldn’t be commenting had I not read the first post, I guess my choice of words was poor, I’ll rephrase. Is there a way to add the repo to hassio, so that it can be installed from within Home Assistant?

Cool! I wish I can use this for facial recognition. Facebox works great but limited on free plan.

1 Like

Great to see the interest in Deepstack :slight_smile: We (HA dev collective) are talking with the Deepstack maintainers about getting a tighter integration with HA, including making it possible to run their containers on a pi. But thats probably a ways of yet owing to technical hurdles that need to be overcome.

@matts1900 thanks for your questions, I will put answers on the Github repo, since otherwise I end up answering the same questions over and over on the forums

3 Likes

Would be interesting to hear your findings on comparing this with tensorflow.

1 Like

Whats important to appreciate is that there is a treadeoff between model size and accuracy - you can have a more accurate model but they will require more RAM and take longer to process. Also tensorflow lets you use any model you want, so I’m not sure what you would be comparing.
Cheers

2 Likes

Thanks for clarifying. You are right, it depends on the use case. I will give it a try on my hardware and will see if it detects persons faster, and if recognition is more precise in terms of confidence. I definetly like the simple fact that it can be run in a separate docker. So it may be a better fit in my case.
I really apreciate your work and your fantastic components I am using on a daily basis!

2 Likes

@BenDiss and all, I just published v0.4 which adds a service deepstack_teach_face so that you can teach (register) faces with Deepstack from the HA UI. Any feedback let me know

2 Likes

When using the face detection (0.4) i get the following error:

Hmm strange. Please open an issue on the Github repo

1 Like

Its working for me. Thanks by the way, I have tried most of your great components. This one is nice and easy and fast i dig it. Im trying to teach it some faces but not having any luck, its probably the path.

I run HA in docker on ubuntu, i have a file located here ‘/home/homeassistant/.homeassistant/www/d2.jpg’

‘.homeasssitant’ is my config folder according to docker, so when i call files in other automations or scripts ‘/local/example.png’ works fine.

Examples:

{
  "name": "Batman",
  "file_path": "/local/d2.jpg"
}
{
  "name": "Batman",
  "file_path": "http://ipaddress:port/local/d2.jpg"  ##Ive tested and this does pull up the image in a browser.
}
{
  "name": "Batman",
  "file_path": "/home/homeasssitant/.homeassistant/d2.jpg"
}

I dont see anything in the logs about it either.

Any help is appreciated!

1 Like