Face and person detection with Deepstack - local and free!

eyalco · January 23, 2020, 5:36pm

Does the hass face component supports google coral? If not, are you familiar with component that does support it (detection / recognition)?

Thx

robmarkcole · January 23, 2020, 5:49pm

Not yet, possible in time

dvbit · January 26, 2020, 3:19pm

Is it possible to use the face component with a local file camera?

robmarkcole · January 26, 2020, 4:08pm

Yes you can use any camera supported by HA

peteonus · January 31, 2020, 8:59pm

@robmarkcole

I’m having trouble with the notification couter, for example if 3 people is detected on the image, I receive the same notification 3 times, i’m using this:

id: ‘1580423474230’
alias: person detection front camera send pushbullet notify
description: ‘’
trigger:
- event_data:
  object: person
  event_type: image_processing.object_detected
  platform: event
  condition: []
  action:
- data:
  message: Person detected on the front door
  service: notify.pushbullet

any help with this?

thanks

robmarkcole · February 1, 2020, 4:25pm

use this approach

Fraschi · February 6, 2020, 6:43pm

Just found that the deepstack url for raspberrypi installation are broken, anyone has a copy of the deepstack.deb file (the one for raspberrypi and ncs)?

xeryax · February 11, 2020, 2:48am

Thanks Rob, for an awesome integration… i’ve been using it for a while now, and it’s time to clean up my config and get everything working as it is supposed to…

That said, I’ve got the following two lines in my config:

    save_file_folder: /tmp/detector/deepstack/
    save_timestamped_file: True

and I am getting filenames like this:

F0DYQO~W.JPG
FC1DYT~S.JPG
FPVETY~G.JPG
FPVETY~L.JPG

Any idea where i may have gone wrong?

robmarkcole · February 11, 2020, 2:12pm

Hmm the component could not produce those filenames

Ekstrom · February 11, 2020, 6:52pm

I am getting those too and it is because :
save_file_folder: /config/snapshots/
save_timestamped_file: True

In windows the names look a little crazy, but if you look inside home assistant its looks normal:

surge919 · February 12, 2020, 3:13am

Am I doing something wrong here?
How do I choose which items trigger notifications?
It’s finding other items but only creating a person detected file.
Also, when I walk in the room, it’s not triggering the automation which should send me the image.
I don’t have the portion of the log below, but when I leave the room, the summary changes from person=1 to person=0
When I enter the room, it changes back to person=1 . but the automation isn’t emailing me anything

My current detection shows a change from
summary=
dining table=1
person=1
laptop=1
chair=2

to
summary=
dining table=2
person=1
laptop=1
chair=3

and the only file that gets created is

deepstack_detector_latest_person.jpg

line of my debug log

2020-02-11 22:03:03 DEBUG (MainThread) [homeassistant.components.websocket_api.http.connection.12334115914] Sending {‘id’: 2, ‘type’: ‘event’, ‘event’: <Event state_changed[L]: entity_id=image_processing.deepstack_detector, old_state=<state image_processing.deepstack_detector=0; last_person_detection=2020-02-11_22:02:20, summary=dining table=1, person=1, laptop=1, chair=2, unit_of_measurement=person, friendly_name=deepstack_detector @ 2020-02-11T22:02:41.917945-05:00>, new_state=<state image_processing.deepstack_detector=1; last_person_detection=2020-02-11_22:03:02, summary=dining table=2, person=1, laptop=1, chair=3, unit_of_measurement=person, friendly_name=deepstack_detector @ 2020-02-11T22:03:03.131442-05:00>>}

configuration.yaml

image_processing:
  - platform: deepstack_object
    ip_address: 10.0.0.20
    port: 5555
    api_key: < my key >
    scan_interval: 20 # Optional, in seconds
    save_file_folder: /config/www/
    save_timestamped_file: True
    source:
    - entity_id: camera.frontdoor
      name: deepstack_detector

automations.yaml

- id: '1159403824611'
  alias: New detection alert
  trigger:
  - platform: event
    event_type: image_processing.object_detected
  action:
  - service: script.send_detected_image

scripts.yaml

send_detected_image:
  sequence:
  - service: notify.gmail
    data_template:
      title: 'Something/Someone was detected'
      message: "What/Who is it?"
      data:
        file: "{{ trigger.event.data.file }}"

I think the issue with my automation not firing is the last line

        file: "{{ trigger.event.data.file }}"

How do I get the filename of the last file created by the automation?

cooloo · February 18, 2020, 9:21am

I’m getting the same over samba in Windows. Its fine in Hass/linux though

robmarkcole · February 22, 2020, 9:07am

It appears that the files with funny names are all being produced on windows platforms only? @cooloo @Ekstrom @xeryax ?

@Dayve67 as per docs:

The box coordinates and the box center (centroid) can be used to determine whether an object falls within a defined region-of-interest (ROI). This can be useful to include/exclude objects by their location in the image.

Basically use the box in a condition on your automation.

@surge919 it probably is a yaml formatting error somewhere

ckold · February 24, 2020, 10:30am

I have not had any false notifications since 2 months but today I got one! I think I just have to raise my confidence level to 85% from the standard. I do wonder however if there are any updates done the the DeepStack docker image that can increase the detection? Is it trained and final or are there regular updates? Thanks for a very good add on to my smart home!

robmarkcole · February 24, 2020, 1:23pm

Deepstack is using Yolo3 which is a standard model. If you want to use your own model checkout the tensorflow integration

ric · February 29, 2020, 6:37am

Hi there, I’m sorry if this has already been answered somewhere but I can’t find anything about. Currently I’m running 2 Docker containers built from the same image, one for face recognition on port 5000 and the other for object recognition on custom port 5005.
So far is working well with 2 cams, but I was wondering what’s the better approach, if any, regarding calculation power between the one you suggest (activating both obj and face on one instance) and mine.
Oh I’m on a i5 NUC 7th gen with 16Gb RAM and M.2 SSD.
Thanks

peteonus · March 9, 2020, 1:47pm

Hi, i’m having trouble this component to trigger an automation when person is detected, the component itself looks like to be working fine because i can see the picture created with the person enclosed in red, here is my automation:

- id: '1234'
  alias: person detection pushbullet notify
  description: ''
  trigger:
  - event_data:
      object: person
    event_type: image_processing.object_detected
    platform: event
  condition: []
  action:
  - delay: '5'
  - data:
      data:
        file: images/deepstack_person_detector_front_latest_person.jpg
      message: Person detected on the front door
      target:
      - channel/mychannel
    service: notify.pushbullet

if i trigger the automation manually i receive the pushbullet message with the picture, but i can’t get the automation to work.

anyone has this issue?

b.maia81 · March 11, 2020, 11:21am

Hey Guys,
I’m thinking about using Deepstack, but need to clarify some ideas.
I have a dell 3060 (win 10 machine) running Blue Iris.
Can I install Deepstack also on this machine, and then after the image is analyzed by Deepstack to send it to Home Assistant? For notification, etc.
Is there a guide for this? My HA is on a Raspbery Pi 3 b+ hass.io.

robmarkcole · March 11, 2020, 3:58pm

@b.maia81 there is a guide here

b.maia81 · March 11, 2020, 5:09pm

Thanks Robin.
If I remember correctly this was the original topic that got me interested in Deepstack. I will give it a deeper look. But my question is more specific, just because I’m completly noob on linux.
Should I make the effort to instal linux on a virtual machine on my DELL pc and instal deepstack on it or windows will also work fine?
This doesn’t mention Blue Iris, but it shouldn’t affect in performance right? the camera feed is handle by BI and then by Deepstack.
Thanks again