[Support]: Wrong detection zones after upgrade to 0.17.0

[B0ndo2]

Describe the problem you are having

I upgraded today to 0.17.0 and made no changes to my configuration at all.
I am getting alerts now showing people walking on the sidewalk and the zone is shows as my driveway. this was working perfectly before

Version

0.17.0-f0d69f7

What browser(s) are you using?

No response

Frigate config file

# frigate.yml
detectors:
  coral:
    type: edgetpu
    device: usb
    #model.path: /media/frigate/my_model_float32.tflite
    # model.path: /media/frigate/my_model_float16_yolov8m_20250401.tflite
    #model.path: /media/frigate/my_model_float16.tflite
model:
  #path: /media/frigate/efficientdet_lite3_512_ptq_edgetpu.tflite
  width: 512
  height: 512
  path: /media/frigate/mobilenetv2_512_8041.tflite
  #width: 640
  #height: 640
  labelmap_path: /media/frigate/coco_labels.txt
  #path: /media/frigate/model_int8_qat.tflite
 #input_dtype: int8
#  cpu1:
#    type: cpu
#logger:
  #logs:
    #frigate.record.maintainer: debug
mqtt:
  # Required: host name
  host: xxxx
  # 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: xxxx
  # 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: false
  # Optional: Width of the output resolution (default: shown below)
  #width: 2560
  # Optional: Height of the output resolution (default: shown below)
  #height: 1440
  # 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:
#  log:
#    exec: trace
#  streams:
#    driveway:
#      - ffmpeg:rtsp://admin:Mohamed21@192.168.6.140/1#video=copy#audio=aac
#      - ffmpeg:rtsp://admin:Mohamed21@192.168.6.140/1
      #- "ffmpeg:driveway#audio=aac"
#      - rtsp://admin:Mohamed21@192.168.6.140/1
#    front_porch_camera:
#      - ffmpeg:rtsp://admin:Mohamed21@192.168.6.141/1#video=copy#audio=aac
    #the next line is to make another copy of the stream with different audio encoding
    #  - "ffmpeg:front_porch_camera#audio=aac"
cameras:
  driveway:
    timestamp_style:
      position: tr
      color:
        red: 255
        green: 0
        blue: 0
    zones:
      drive_way:
        coordinates: 
          0.406,0.152,0.49,0.152,0.635,0.162,0.741,0.197,0.768,0.258,0.817,0.552,0.856,0.76,0.857,0.821,0.774,0.909,0.676,0.987,0.246,1,0.257,0.355,0.284,0.171
        objects:
          - person
          - car
        loitering_time: 0
        inertia: 3
      apron:
        coordinates: 0.612,0.09,0.59,0.045,0.38,0.047,0.363,0.076,0.515,0.076
        inertia: 3
        loitering_time: 0
        objects: car
      sidewalk:
        coordinates: 
          0.21,0.139,0.37,0.082,0.477,0.076,0.595,0.092,0.616,0.101,0.756,0.141,0.758,0.177,0.666,0.156,0.613,0.151,0.561,0.147,0.483,0.142,0.399,0.141,0.225,0.173,0.179,0.191
        inertia: 2
        loitering_time: 1
        objects: car
      front_yard:
        coordinates: 
          0.747,0.197,0.769,0.254,0.818,0.54,0.862,0.758,0.974,0.606,0.837,0.24
        #inertia: 1
        loitering_time: 0
        objects: person
        inertia: 1
    objects:
      track:
        - person
        - car
      filters:
        person:
          #mask: 
            #0.322,0.133,0.456,0.125,0.489,0.132,0.521,0.135,0.555,0.137,0.597,0.14,0.62,0.138,0.637,0.142,0.625,0.111,0.552,0.083,0.475,0.079,0.441,0.087,0.214,0.129,0.18,0.173
          min_score: 0.6
          threshold: 0.6
          min_area: 1300
          max_area: 120000
        car:
          min_area: 5000
          max_area: 300000
          min_score: 0.55
          threshold: 0.65

    ffmpeg:
      output_args:
        record: preset-record-generic-audio-aac
      inputs:
        - path: rtsp://xxx@192.168.6.xx/1
        #- path: rtsp://127.0.0.1:8554/driveway
          roles:
            - detect
            - audio
            - record
     #   - path: rtsp://xxx@192.168.6.xx/1
          #input_args: preset-rtsp-restream
      #    roles:
      #      - record

    audio:
      enabled: false
      min_volume: 600
      listen:
        - scream
        - speech
        - yell
    detect:
      width: 1280
      height: 720
#      fps: 5
      #stationary:
        #interval: 0
      #  threshold: 50
    motion:
      mask:
        - 0,0,0.283,0,0.156,0.225,0.068,0.499,0.009,0.869,0,0.861
        - 1,0,0.739,0,0.764,0.181,0.892,0.278,1,0.531
        - 0.366,0.039,0.597,0.05,0.628,0.071,0.69,0.082,0.755,0.136,0.741,0,0.284,0,0.241,0.076
        - 1,0.677,0.728,0.993,0.998,0.998
      threshold: 63
      contour_area: 60
      frame_height: 150
      lightning_threshold: 0.7
      improve_contrast: true
    record:
      enabled: true
      alerts:
        retain:
          days: 7
      detections:
        retain:
          days: 1
    snapshots:
      enabled: true
      bounding_box: true
      required_zones:
        - drive_way
        - apron
        - front_yard
      timestamp: true
      retain:
        default: 7
    review:
      alerts:
        labels:
          - car
          - person
        required_zones:
          - drive_way
          - apron
          - front_yard
      detections: {}
  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://xx@192.168.6.xx/2
          roles:
            - detect
        - path: rtsp://xx@192.168.6.xx/1
          #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
      threshold: 43
      contour_area: 30
      improve_contrast: true
    record:
      enabled: true
      alerts:
        retain:
          days: 5
      detections:
        retain:
          days: 5
    snapshots:
      enabled: true
      bounding_box: true
      timestamp: true
      retain:
        default: 7

version: 0.17-0
camera_groups:
  Home:
    order: 1
    icon: LuCamera
    cameras: birdseye
detect:
  enabled: true

Relevant Frigate log output

N/A

Relevant go2rtc log output

N/A

FFprobe output from your camera

N/A

Frigate stats

No response

Install method

Home Assistant Add-on

docker-compose file or Docker CLI command

N/A

Object Detector

Coral

Network connection

Wireless

Camera make and model

N/A

Screenshots of the Frigate UI's System metrics pages

No response

Any other information that may be helpful

No response

[inkeep-ai-assistant[bot]]

The zone detection behavior you're describing is working as designed in Frigate. Zone presence is evaluated based on the bottom center of the bounding box for the object(1).

Looking at your configuration, you have a sidewalk zone defined with loitering_time: 1, but this zone is not included in your required_zones for alerts(2). Your alert configuration only requires these zones:

review:
  alerts:
    required_zones:
      - drive_way
      - apron
      - front_yard

(2)

Since the sidewalk zone isn't in the required_zones list for alerts, objects detected in the sidewalk zone could still trigger alerts if their bottom center point enters one of your required zones (drive_way, apron, or front_yard) at any point during tracking(1).

To restrict alerts to only occur when objects are in specific zones, you need to ensure your required_zones configuration includes all the zones where you want alerts(3). If you want to prevent sidewalk detections from creating alerts, consider:

1. Adding the sidewalk zone to required_zones if you want alerts from that zone
2. Or adjusting your zone coordinates so the sidewalk area doesn't overlap with your drive_way, apron, or front_yard zones
3. Using zone inertia to require the object's bounding box to be within the zone for multiple consecutive frames before being considered present(1)

The zone inertia feature requires an object's bounding box to be within the zone for the specified number of consecutive frames (default is 3) before the object is considered present in that zone(1)(2).

📚 Sources:

1. Zones
2. Full Reference Config
3. Review

---

Have another question? Just tag @inkeep.

[B0ndo2]

This configuration worked fine with the previous version and I had no wrong alerts so something must have changed in this release

[hawkeye217]

No, nothing has changed with object tracking in 0.17.

If you click on the point that appears in the orange zone, it should show you the bounding box that your model detected. If I had to guess, it's because the model you're using interpreted the two walkers and the dog as a single person.

[B0ndo2]

I haven't changed the model am using between upgrades. I only clicked update and started Frigate.
My alert is now full of similar cases that used to go to detection and I don't know what to do.

[hawkeye217]

Can you post a screenshot of what the bounding box looks like when you click on the dot in the orange zone?

I'd suggest trying YOLOv9 with your Coral, it's a much more accurate model than the one you're currently using. See the official docs: https://docs.frigate.video/configuration/object_detectors/#yolov9

[B0ndo2]

I couldn't get the bounding box on this exact point, this is the best I could do (if I click on the orange dot, the bounding box disappears).
I am using a custom trained model which was working with no issues with the previous release.

[B0ndo2]

Same model but new release, things are acting weird

[B0ndo2]

Frigate makes config and DB backups automatically which should be in the addon specific folder where the config is, I'm not aware of any possibility to choose a version of an addon that is not the latest, but maybe it exists.

I have a copy of the config file and the DB but I don't know how to get an older version of Frigate in HA

[B0ndo2]

I was able to restore the integration (now called App) to 0.16.1 and I copied back frigate.db and config.yml.
The problem is that old recordings are not visible in Frigate UI.
I updated on March 2, my retention is 7 days but they are not visible and I am guessing will not be deleted when the time comes

[hawkeye217]

You can use the sync_recordings option to clean up orphaned recordings.

[B0ndo2]

The problem is likely the change in the way larger regions are handled. 512 is an odd region size and this leads to Frigate using regions down to 256x256 pixels depending on motion

My recommendation would be to maybe try clearing out the regions DB, otherwise we can adjust the logic slightly as I had not considered mid-sized models since 320 and 640 are the common sizes

Is this fixed in the latest release (0.17.1) ?

[NickM-27]

Yes, now it sends at minimum 320x320 region and won't scale up from lower than that

[B0ndo2]

I will test it soon and report back

[NickM-27]

Looking again and it looks like this change might not have made it in to that release.

[B0ndo2]

When do you expect to get the fix out ?

[NickM-27]

We currently have no plans for 0.17.2, but if that changes then it would come then. Otherwise it will be released as part of 0.18 which is currently in testing / pre-beta phase