Frigate False Positives object detection

Hello all,

I have frigate NVR setup on my Home assistant OS in a VM.
I run frigate locally in my HASSOS virtual machine.
I recently bought a google coral stick to help with the fps but having issues with that but thats besides the point I have a different topic on the forum relating to that.
What I want to ask is what can I do to help stop getting so many false positives. I set up some automations that send me alerts when a Person is in my yard, etc. and I get notifications of chairs/tiki torches/ my mailbox, random objects like my grill, and so many more that come ALL the time saying that all of these are PEOPLE.
Here are some examples:
I would love to do anything to help with this in the settings, etc. I want to sit down and test a minimum area and see what the sizes are to maybe set one at different zones in my camera to see if that would help.
garbagecan
mailbox
FalseChairPerson
FalseGrillPerson
FalseTablePerson
FalseTikiPerson

Here is my Frigate.yml

detect:
  width: 1920
  height: 1080
  fps: 7
  stationary:
    interval: 10 
objects:
  track:
    - person
    - car
    - dog
  #mask: stops all objects from being tracked here
  filters:
    person:
      min_score: 0.4 # min score for object to initiate tracking (default: 0.5
      threshold: 0.55 # min decimal percentage for tracked object's computed score to be considered a true positive (default: 0.7)
    car:
      threshold: 0.6
record:
  enabled: True
  retain:
    days: 3
    mode: active_objects
  events:
    retain:
      default: 3
      mode: active_objects
      objects:
        person: 10
cameras:
  backyard:
    ffmpeg:
      inputs:
        - path: rtsp://10.0.0.14:554/11
          roles: 
            - detect
            - record
            - rtmp
    objects:
      track:
        - person
        - dog
        - cat
        - car
      filters:
        person:
          mask:
            - 293,306,307,273,348,250,428,229,656,115,733,227,817,177,734,98,710,71,988,0,0,0,0,389,35,372,79,349,134,353,159,387
            - 1920,961,1920,1080,1797,1042,1776,968,1795,923,1806,853,1745,840,1657,815,1575,978,1491,952,1608,795,1665,675,1714,584,1776,466,1893,346,1920,348
    zones:
      back_alley:
        coordinates: 547,432,230,637,185,557,110,424,261,324,428,230,735,78,784,213
        objects:
          - car
          - person
        # filters:
        #   person:
        #     # min_area: 5000
        #     # max_area: 100000
        #     # threshold: 0.7
        # Optional: Zone level object filters.
        filters:
          person:
            threshold: 0.4
            min_score: 0.4
      back_porch:
        coordinates: 1920,1080,1920,873,1625,815,1674,681,1204,548,562,578,381,512,0,790,0,1080
        objects:
          - person
          - dog
          - cat
        filters:
          person:
            min_area: 5000
            max_area: 100000
            threshold: 0.7
          dog:
            min_area: 1000
            max_area: 75000
            threshold: 0.6
  frontyard:
    ffmpeg:
      inputs:
        - path: rtsp://10.0.0.11:554/11
          roles: 
            - detect
            - record
    objects:
      track:
        - person
        - car
        - dog
      filters:
        person:
          threshold: 0.7
    zones:
      front_parking:
        coordinates: 316,150,413,147,780,257,1047,359,1331,458,1548,548,1669,605,1774,678,1328,509
        objects:
          - car
          - person
      frontzone:
        coordinates: 0,1080,0,360,250,223,362,249,600,348,924,461,1194,550,1920,826,1920,1080
    record:
      enabled: true
      retain: 
        days: 2
        mode: active_objects
      events:
        retain:
          default: 3
          mode: motion #active_objects

You are of course going to get a lot of person false positives with min_score and threshold that low. Definitely would recommend keeping that at the normal level.

For objects like the mailbox or chair which are in the same place, a small object mask at the bottom center of the object would be recommended.

For other things, min_area and max_area filters should work well too

How would I do a small object mask at the bottom center of the object?
Do you mean just put a mask over the object like my mailbox?
My chairs are randomly moved all the time from my kids lol so that wouldn’t work on that one.
I’ll increase the min_score I forgot I did that (lowering it) when I was testing months ago.