Dog is detected as a person

[SmartM-ui]

Hi @blakeblackshear , why even if I have configured the detection of people only, when my dog passes in front of the camera I still receive the detection notice in person? He is recognized as a person at 65%, sometimes even when I pass in front of the camera he recognizes me as a person at 65%!

Can I improve something?

    detect:
      width: 1920 
      height: 1080
      fps: 5
      enabled: True
      max_disappeared: 25

     record:
       enabled: True
       retain_days: 0
       events:
         pre_capture: 5
         post_capture: 5
         objects:
           - person
         required_zones: []
         retain:
           default: 10
           objects:
             person: 10

     # Optional: Camera level object filters config.
     objects:
       track:
         - person
       filters:
         person:
           min_area: 5000
           max_area: 100000
           min_score: 0.4
           threshold: 0.5

(the human detection of the webcam works correctly, in fact the official app does not detect the movement of the dog)

Thanks

[blakeblackshear]

This just happens sometimes with the model provided by Google. Until I develop a better model, it's expected behavior.

[SmartM-ui]

This just happens sometimes with the model provided by Google. Until I develop a better model, it's expected behavior.

Hi @blakeblackshear

Would it be impossible to predict the use of the human recognition sensor for a single camera, bypassing the google system?

It would be useful (at least for me) to continue using the frigate system (really great), but to record the video only when the human sensor of the cam switches to the ON state (Only for the start of the video recording)

Thanks as always for the quick answers you give

[blakeblackshear]

Sounds like you are confused about how frigate works. The AI models use tensorflow and are trained and provided by Google. It all runs locally, but the technology is open source by Google. This is how human detection works. Sometimes it thinks a dog looks like a human.

[SmartM-ui]

Sounds like you are confused about how frigate works. The AI models use tensorflow and are trained and provided by Google. It all runs locally, but the technology is open source by Google. This is how human detection works. Sometimes it thinks a dog looks like a human.

I more or less know how Frigate works :-)

It was to have an extra option, in the sense, to continue to exploit the functionality of the NVR, but in some cases to use human recognition not from Google, but directly from the Cams.

Sorry again for the request

However the strange thing is that I have two cameras, on two floors of the house, both use the same frigate and have a similar configuration (change the resolution). When the dog gets on and off the floor, the higher resolution cam mistakes it for human and sends the notification, the lower resolution cam works correctly and does not confuse it with a human.

[blakeblackshear]

There will be a way to externally trigger an event in future versions. I can't speak to your cameras ability to integrate via mqtt or http.

[SmartM-ui]

There will be a way to externally trigger an event in future versions. I can't speak to your cameras ability to integrate via mqtt or http.

Thanks! For info, my cam is integrated into HA via MQTT

[joaofgf]

I have the same problem, different animal...