New image_processing custom component - HASS-Google-Vision

Hi @ros
this is a hot topic in research, and doing anything fancy is probably going to require a lot of thought and effort. For an overview of the topic checkout this article. My advice is not to leap into doing anything involving bayesian sensor, but to start as simple as possible. For example if one model is doing well in daytime, and another in low light, then use a simple time based rule to determine which model to use. In the meantime as you develop more sophisticated rules, keep a log of model results and try to spot any patterns that will be useful, e.g. which model does well and when. This would be very interesting to see and help guide development for people like myself who spend more time developing and not enough time hands on with the tools.
Cheers

1 Like

Can we use this system to recognize people in my household or this is just generic persons?

Just generic. For recognising faces locally try dlib or facebox

1 Like

To use it, do I need to do all the steps from https://www.home-assistant.io/integrations/google_cloud/? That is: enable billing for project? I still get error:

Cloud Vision API has not been used in project XXXXX before or it is disabled

I did enable it and still error :frowning:

Yeah getting the settings right on GCP can be confusing. If you get tired trying tryout Sighthound

I did but Sighthound is less accurate. Lately it took flower as person :wink: I sent the very same image to GCP and no mistake. But it has one big advantage over GCP -> no need to set billing card…

Yes that is an advantage :slight_smile:
All of the services will make mistakes sometimes, and they are also probably making a different tradeoff of speed vs accuracy

Hi @robmarkcole , thanks for another great component. been lookin for this for a while. Working great except that one issue. sometime it recognises me as “man” but not " person". hence not triggering the automation. Is not man a person? (no pun intended)

Google choose the labels - I noticed that and its a bit wierd. Please create a feature request on the repo explaining the solution

1 Like

anyone able to get the targets attribute working?

have tried

target:
      - person
      - car
      - wildlife

but get this:

Invalid config for [image_processing.google_vision]: value should be a string for dictionary value @ data['target']. Got ['person', 'car', 'wildlife']. (See ?, line ?).```

Only 1 target is possible with this integration. Chechout my amazon rekognition integration instead

Hi Can you say me what kind of camera do you use ?
Regards
Giloris

Use any kind of camera, personally I am using RPi cameras

Hi Thank you. Your custom component is top. I build a chicken coop connected and i want counting eggs produced every days with camera .
But i use ESP32 Cam and i dont know if it’s compatibility.
How google vision find it ? i dont see anywhere on program ID camera use ?
Regards

You need your ESP32 Cam as a camera in home assistant.

Hi
How declare it on home assistant ? I need an IP camera only , i suppose.
camera integration ?. rpi camera need connected to raspberry module but module is my home and camera is outdoor !
Regards
Giloris

@giloris checkout esphome camera, note this is now off topic (HASS-Google-Vision)

Hi
Sorry
I will try finding information to another help forum
Regards
Giloris

1 Like

Hi, I have instal last home assistant version (0.115.2) and google vision dont work now !
I have thes errors lines :

  Platform error image_processing.google_vision - Error relocating /usr/local/lib/python3.8/site-packages/grpc/_cython/cygrpc.cpython-38-arm-linux-gnueabihf.so: backtrace: symbol not found

Have you an idea ?
Thanks
Giloris

Appears a dependency got upgraded somewhere breaking the integration.
Sorry but I have too many projects to maintain and google vision is not a high priority, so unless someone wants to jump in with a fix I will archive this project on 1st Oct