Person not detected (multiple times)

[B0ndo2]

Describe the problem you are having

The person in the photo was not detected on the driveway in more than one instance (over a few days).

There are no errors in the log file, other people are detected right before this person so I feel like it has to do with the colours/shape of the person as it relates to the model ?

Version

0.14.1-f4f3cfa

What browser(s) are you using?

No response

Frigate config file

# frigate.yml
detectors:
  coral:
    type: edgetpu
    device: usb
#  cpu1:
#    type: cpu
logger:
  logs:
    frigate.record.maintainer: debug
mqtt:
  # Required: host name
  host: caha.duckdns.org
  # Optional: port (default: shown below)
  port: 1883
  # Optional: topic prefix (default: shown below)
  # WARNING: must be unique if you are running multiple instances
  topic_prefix: frigate
  # Optional: client id (default: shown below)
  # WARNING: must be unique if you are running multiple instances
  client_id: frigate
  # Optional: user
  user: frigate
  # Optional: password
  # NOTE: Environment variables that begin with 'FRIGATE_' may be referenced in {}.
  #       eg. password: '{FRIGATE_MQTT_PASSWORD}'
  password: ssss
  # Optional: interval in seconds for publishing stats (default: shown below)
  stats_interval: 60
motion:
#  improve_contrast: false
  contour_area: 30
  threshold: 40
birdseye:
  # Optional: Enable birdseye view (default: shown below)
  enabled: true
  # Optional: Width of the output resolution (default: shown below)
  width: 1280
  # Optional: Height of the output resolution (default: shown below)
  height: 720
  # Optional: Encoding quality of the mpeg1 feed (default: shown below)
  # 1 is the highest quality, and 31 is the lowest. Lower quality feeds utilize less CPU resources.
  quality: 8
  # Optional: Mode of the view. Available options are: objects, motion, and continuous
  #   objects - cameras are included if they have had a tracked object within the last 30 seconds
  #   motion - cameras are included if motion was detected in the last 30 seconds
  #   continuous - all cameras are included always
  mode: continuous
go2rtc:
  streams:
    driveway:
      - rtsp://admin:XXXX@192.168.6.140/1
    front_porch_camera:
      - rtsp://admin:XXXX@192.168.6.141/1

cameras:
  driveway:
    timestamp_style:
      position: tr
      color:
        red: 255
        green: 0
        blue: 0
    zones:
      sidewalk:
        coordinates: 409,52,368,17,241,18,227,50
        #coordinates: 114,64,232,57,302,62,409,68,430,50,480,62,521,80,553,84,529,51,389,20,242,16,176,25,147,34
        objects:
          - car
      drive_way:
        coordinates: 274,47,441,62,508,76,562,189,550,260,562,286,429,352,0,352,30,207,108,66
        objects:
          - person

    objects:
      track:
        - person
        - car
      filters:
        person:
          mask:
            - 114,64,231,43,292,43,364,47,430,50,480,62,521,80,553,84,529,51,389,20,242,16,176,25,147,34
          threshold: 0.6
          min_area: 1300
          max_area: 25000
    ffmpeg:
      inputs:
        - path: rtsp://admin:XXXX@192.168.6.140/2
          roles:
            - detect
            #- audio
        - path: rtsp://127.0.0.1:8554/driveway
          input_args: preset-rtsp-restream
          roles:
            - record

    audio:
      enabled: false
      min_volume: 600
      listen:
        - scream
        - speech
        - yell
    detect:
      width: 640
      height: 360
      fps: 5
   #   stationary:
    #    interval: 0
    #    threshold: 50
    motion:
      mask:
        - 0,0,0.283,0,0.156,0.225,0,0.844
        - 1,0,0.739,0,0.764,0.181,0.892,0.278,1,0.531
        - 0.366,0.039,0.623,0.05,0.787,0.125,0.741,0,0.269,0,0.236,0.075
        - 1,0.677,0.728,0.993,0.998,0.998
      threshold: 30
      contour_area: 26
      improve_contrast: 'true'
    record:
      enabled: true
      retain:
        days: 0
      events:
        retain:
          default: 5
          mode: active_objects
    snapshots:
      enabled: true
      bounding_box: true
      required_zones:
        - drive_way
        - sidewalk
      timestamp: true
      retain:
        default: 7


################################################ 
    review:
      alerts:
        required_zones:
          - drive_way
          - sidewalk
  front_porch_camera:
    objects:
      track:
        - person
      filters:
        person:
          #threshold: 0.75
          mask:
            - 176,352,171,122,173,12,377,8,518,57,640,168,640,0,0,0,0,352
    ffmpeg:
      output_args:
        record: preset-record-generic-audio-aac
      inputs:
        - path: rtsp://admin:XXXX@192.168.6.141/2
          roles:
            - detect
        - path: rtsp://127.0.0.1:8554/front_porch_camera
          input_args: preset-rtsp-restream
          roles:
            - record
            - audio
  
    
    audio:
      enabled: true
      min_volume: 600

    detect:
      width: 640
      height: 360
      fps: 5
    motion:
      mask:
        - 176,352,171,122,173,12,377,8,518,57,640,80,640,0,0,0,0,352
    record:
      enabled: true
      retain:
        days: 0
      events:
        retain:
          default: 5
          mode: motion
    snapshots:
      enabled: true
      bounding_box: true
      timestamp: true
      retain:
        default: 7

version: 0.14
camera_groups:
  Home:
    order: 1
    icon: LuCamera
    cameras:
      - driveway
      - front_porch_camera

Relevant Frigate log output

NA

Relevant go2rtc log output

NA

FFprobe output from your camera

NA

Frigate stats

No response

Install method

HassOS Addon

docker-compose file or Docker CLI command

NA

Object Detector

Coral

Network connection

Mixed

Camera make and model

Chortau

Screenshots of the Frigate UI's System metrics pages

No response

Any other information that may be helpful

No response

[NickM-27]

Remember that the default model is trained on the COCO dataset which is not using security camera images, this is especially prevalent for people from the top-down perspective. A recommendation here would be using a better model, like Frigate+, which is trained on images like this and allow you to improve the model yourself.

[B0ndo2]

You mean use the high res. stream for detection?

[B0ndo2]

@NickM-27 Can you please clarify ?

[B0ndo2]

I was using the low res stream which is 640x360 since the documentation said that higher resolution will not benefit detection. Did I misunderstand?
https://docs.frigate.video/frigate/camera_setup/#choosing-a-detect-resolution

[NickM-27]

Higher resolution in the documentation means truly high resolution. 1280x720 is generally considered the sweet spot

[B0ndo2]

Unfortunately my second stream is 1920x1080. I am using it now and using 1280x720 for detect but ffmpeg is using 22% of my vm cpu which is ok I guess

[B0ndo2]

Do you believe that my suspicions to why this person was not detected is true ?

[NickM-27]

Not sure I know what you are asking, I already explained what the issue likely which seems to align with your understanding. A lot more information could be gathered by watching the live debug view so we can see more accurately what happened. Perhaps the person was detected just with a lower score.