search

blakeblackshear / frigate

NVR with realtime local object detection for IP cameras
https://frigate.video
MIT License
19.21k stars 1.76k forks source link

[Support]: current_zones MQTT object sometimes empty #4637

Closed J-Pipe closed 1 year ago

J-Pipe commented 1 year ago

Describe the problem you are having

I'm logging the MQTT messages of the events, and sometimes the "current_zones" object is empty even when the object is in a zone that should detect that object.

Version

0.11.1-2eada21

Frigate config file

# yaml-language-server: $schema=http://10.0.1.62:5000/api/config/schema

#Example config here:
#https://docs.frigate.video/configuration/index/

mqtt:
  host: 10.0.1.62
  user: 
  password: 

detectors:
  coral:
    type: edgetpu
    device: usb

#https://docs.frigate.video/configuration/camera_specific
# Reference - Blue Iris RTSP Cameras
ffmpeg:
  # todo - see if removing some of this helps playback
  hwaccel_args:
    - -hwaccel
    - vaapi
    - -hwaccel_device
    - /dev/dri/renderD128
    - -hwaccel_output_format
    - yuv420p
  input_args: -avoid_negative_ts make_zero -flags low_delay -strict experimental -fflags +genpts+discardcorrupt -rtsp_transport tcp -timeout 5000000 -use_wallclock_as_timestamps 1

rtmp:
  enabled: True # <-- RTMP should be disabled if your stream is not H264

snapshots:
  enabled: True
  clean_copy: False
  crop: True
  required_zones:
    - sidewalk_zone
    - driveway_car_zone
    - driveway_person_zone
    - porch_zone
    - yard_zone
  bounding_box: True
  retain:
  # Required: Default retention days (default: shown below)
    default: 1

record:
  enabled: False
  events:
    objects:
      - person
      - car
      - dog
      - cat
    required_zones:
      - sidewalk_zone
      - driveway_car_zone
      - driveway_person_zone
      - porch_zone
      - yard_zone
      #- street_zone
    retain:
      default: 5
      mode: active_objects

objects:
  # Optional: list of objects to track from labelmap.txt (default: shown below)
  track:
      - person

detect:
  enabled: True
  width: 1280
  height: 720

cameras:
  baby_cam:
    ffmpeg:      
      inputs:
        - path: rtsp://10.0.1.35:3765/eufy_cam
        #- path: /config/www/debug/driveway_sub.mp4 # debug only
        #  input_args: -re -stream_loop -1 -fflags +genpts # debug only
          roles:
            - detect 
    detect:
      width: 1920
      height: 1080
      fps: 1
    objects:
      track:
        - person
    mqtt:
      enabled: False

  driveway_cam:
    ffmpeg:      
      inputs:
        - path: rtsp://10.0.1.35:3765/driveway_sub
        #- path: /config/www/debug/driveway_sub.mp4 # debug only
        #  input_args: -re -stream_loop -1 -fflags +genpts # debug only
          roles:
            - detect
            - rtmp
            - record    

    objects:
      track:
        - person
        - car
      # NOTE: This mask is COMBINED with the object type specific mask below
      mask:
        - 1280,720,1280,320,1249,381,1209,384,1190,421,1204,451,1200,525,1160,720
        - 1061,0,1280,0,1280,343,1160,283,1066,285,992,259,943,208,932,167,765,135,506,77,494,0
        - 20,207,45,207,44,188,20,188
      filters:
        car:
          mask: 
            - 820,215,983,248,1082,163,1280,322,1280,572,1027,414,920,419,655,297,546,198

    motion:
      mask:
        - 976,358,1160,411,1140,471,1167,502,1041,720,1280,720,1280,474,1280,0,0,0,0,27,180,28,491,56,518,111,517,193,563,220,641,233,709,263,779,270,834,293,899,327

    zones:
      # Required: name of the zone
      # NOTE: This must be different than any camera names, but can match with another zone on another
      #       camera.
      sidewalk_zone:
        # Required: List of x,y coordinates to define the polygon of the zone.
        # NOTE: Presence in a zone is evaluated only based on the bottom center of the objects bounding box.
        coordinates: 461,278,762,271,756,204,0,188,0,300
        # Optional: List of objects that can trigger this zone (default: all tracked objects)
        objects:
          - person
        # Optional: Zone level object filters.
        # NOTE: The global and camera filters are applied upstream.
        filters:
          person:
            min_area: 4225
            max_area: 29000
            min_score: 0.5
            threshold: 0.6
            # Optional: mask to prevent this object type from being detected in certain areas (default: no mask)
            # Checks based on the bottom center of the bounding box of the object
            #mask: 0,0,0,0,0,0
      driveway_car_zone:
        coordinates: 177,720,1016,720,1059,683,1042,662,1115,525,843,403,690,354,527,244,343,246,155,254
        objects:
          - car
        filters:
          car:
            min_area: 30000
            max_area: 500000
            min_score: 0.5
            threshold: 0.6
      street_zone:          
        coordinates: 418,282,559,278,527,195,951,226,941,179,782,140,483,96,195,80,0,84,0,189,144,181,154,295,257,290
        objects:
          - car
          - person
        filters:
          car:
            min_area: 1800
            max_area: 755000
            min_score: 0.5
            threshold: 0.6
          person:
            min_area: 600
            max_area: 15000
            min_score: 0.5
            threshold: 0.6
      driveway_person_zone:
        objects:
          - person
        coordinates: 0,720,1108,720,1200,525,1193,481,987,340,791,294,460,282,0,312
        filters:
          person:
            min_area: 7000
            max_area: 110000
            min_score: 0.5
            threshold: 0.6
      #no_alert_zone:
      #  coordinates: 1100,720,1170,540,1056,475,773,481,633,522,868,720

  front_yard_cam:
    ffmpeg:
      inputs:
        - path: rtsp://10.0.1.35:3765/front_yard_sub        
        #- path: /config/www/debug/front_yard_sub.mp4 # debug only
        #  input_args: -re -stream_loop -1 -fflags +genpts # debug only
          roles:
            - detect
            - rtmp
            - record

    objects:
      mask:
        - 1280,70,1280,0,1077,0,0,0,0,584,396,342,326,53,667,34,723,96
    motion:
      mask:
        - 1065,691,789,666,579,633,798,321,629,297,218,666,0,670,0,720,1136,720
        - 0,0,1280,0,1280,106,908,113,0,125

    zones:
      # Required: name of the zone
      # NOTE: This must be different than any camera names, but can match with another zone on another
      #       camera.
      yard_zone:
        # Required: List of x,y coordinates to define the polygon of the zone.
        # NOTE: Presence in a zone is evaluated only based on the bottom center of the objects bounding box.
        coordinates: 1280,720,1280,222,515,219,383,223,386,279,413,345,492,355,660,389,287,720
        # Optional: List of objects that can trigger this zone (default: all tracked objects)
        objects:
          - person
          - cat
          - dog
        # Optional: Zone level object filters.
        # NOTE: The global and camera filters are applied upstream.
        filters:
          person:
            min_area: 8000
            max_area: 120000
            min_score: 0.5
            threshold: 0.6
            # Optional: mask to prevent this object type from being detected in certain areas (default: no mask)
            # Checks based on the bottom center of the bounding box of the object
            #mask: 0,0,0,0,0,0
      sidewalk_zone:
        coordinates: 1280,212,941,212,577,212,613,155,843,142,1280,139
        objects:
          - person
        filters:
          person:
            min_area: 4200
            max_area: 19500
            min_score: 0.5
            threshold: 0.6
      porch_zone:
        coordinates: 285,720,609,410,404,352,0,557,0,720
        objects:
          - person
          - cat
          - dog
        # Optional: Zone level object filters.
        # NOTE: The global and camera filters are applied upstream.
        filters:
          person:
            min_area: 10000
            max_area: 120000
            min_score: 0.5
            threshold: 0.6
     # driveway_zone:
      #  coordinates: 316,203,360,354,527,342,654,310,726,265,775,205,531,195
        #objects: # should use the default tracked objects above
          #- car
     # street_zone:
      #  coordinates: 569,171,766,162,1280,156,1280,64,428,105,360,120,347,144,358,167,408,165
       # objects:
         # - car

  doorbell_cam:
    ffmpeg:
      inputs:
        - path: rtsp://10.0.1.35:3765/doorbell_sub        
        #- path: /config/www/debug/doorbell_sub.mp4 # debug only
        #  input_args: -re -stream_loop -1 -fflags +genpts # debug only
          roles:
            - detect
            - rtmp
            - record

    motion:
      mask:
        - 0,1,0,93,561,167,658,266,619,317,619,415,700,428,682,720,1280,720,1280,0,2,3

    # testing this recommended config for double take (https://github.com/jakowenko/double-take)
    mqtt:
      timestamp: False
      bounding_box: False
      crop: True
      quality: 100
      height: 500

    zones:
      # Required: name of the zone
      # NOTE: This must be different than any camera names, but can match with another zone on another
      #       camera.
      porch_zone:
        # Required: List of x,y coordinates to define the polygon of the zone.
        # NOTE: Presence in a zone is evaluated only based on the bottom center of the objects bounding box.
        coordinates: 0,720,983,720,1035,544,541,496,428,452,0,485
        # Optional: List of objects that can trigger this zone (default: all tracked objects)
        objects:
          - person
          - cat
          - dog
        # Optional: Zone level object filters.
        # NOTE: The global and camera filters are applied upstream.
        #filters:
        #  person:
        #    min_area: 8000
        #    max_area: 120000
        #    min_score: 0.5
        #    threshold: 0.6

#logger:
#  default: debug

Relevant log output

Looking at the "before" objects, the first and third events have empty "current_zone" objects. In the first event, the "current_zones" should include the "driveway_car_zone" zone, and in the third event the "current_zones" should include the "street_zone" zone.

{"before":{"id":"1670460861.454207-v45ue5","camera":"driveway_cam","frame_time":1670460861.454207,"snapshot_time":0,"label":"car","sub_label":null,"top_score":0,"false_positive":true, "start_time":1670460861.454207,"end_time":null,"score":0.57421875,"box":[435,369,1074,708],"area":216621,"ratio":1.8849557522123894,"region":[246,0,1278,1032],"stationary":false,"motionless_count":0,"position_changes":0,"current_zones":[],"entered_zones":[],"has_clip":false,"has_snapshot":false},"after":{"id":"1670460861.454207-v45ue5","camera":"driveway_cam","frame_time":1670460862.198257,"snapshot_time":1670460862.198257,"label":"car","sub_label":null,"top_score":0.74609375,"false_positive":false,"start_time":1670460861.454207,"end_time":null,"score":0.76171875,"box":[358,313,988,718],"area":255150,"ratio":1.5555555555555556,"region":[20,0,1260,1240],"stationary":false,"motionless_count":3,"position_changes":1,"current_zones":["driveway_car_zone"],"entered_zones":["driveway_car_zone"],"has_clip":false,"has_snapshot":true},"type":"new"}
{"before":{"id":"1670460861.454207-v45ue5","camera":"driveway_cam","frame_time":1670460862.198257,"snapshot_time":1670460862.198257,"label":"car","sub_label":null,"top_score":0.74609375,"false_positive":false,"start_time":1670460861.454207,"end_time":null,"score":0.76171875,"box":[358,313,988,718],"area":255150,"ratio":1.5555555555555556,"region":[20,0,1260,1240],"stationary":false,"motionless_count":3,"position_changes":1,"current_zones":["driveway_car_zone"],"entered_zones":["driveway_car_zone"],"has_clip":false,"has_snapshot":true}, "after":{"id":"1670460861.454207-v45ue5","camera":"driveway_cam","frame_time":1670460868.936844,"snapshot_time":1670460862.198257,"label":"car","sub_label":null,"top_score":0.74609375,"false_positive":false,"start_time":1670460861.454207,"end_time":null,"score":0.57421875,"box":[237,97,361,244], "area":18228,"ratio":0.8435374149659864,"region":[0,0,1392,1392],"stationary":false,"motionless_count":0,"position_changes":1,"current_zones":[],"entered_zones":["driveway_car_zone"],"has_clip":false,"has_snapshot":true},"type":"update"}
{"before":{"id":"1670460861.454207-v45ue5","camera":"driveway_cam","frame_time":1670460868.936844,"snapshot_time":1670460862.198257,"label":"car","sub_label":null,"top_score":0.74609375,"false_positive":false,"start_time":1670460861.454207,"end_time":null,"score":0.57421875,"box":[237,97,361,244],"area":18228, "ratio":0.8435374149659864,"region":[0,0,1392,1392],"stationary":false,"motionless_count":0,"position_changes":1,"current_zones":[],"entered_zones":["driveway_car_zone"],"has_clip":false,"has_snapshot":true}, "after":{"id":"1670460861.454207-v45ue5","camera":"driveway_cam","frame_time":1670460875.598356,"snapshot_time":1670460862.198257,"label":"car","sub_label":null,"top_score":0.74609375,"false_positive":false,"start_time":1670460861.454207,"end_time":null,"score":0.71484375,"box":[107,30,405,139], "area":32482,"ratio":2.7339449541284404,"region":[0,0,604,604],"stationary":false,"motionless_count":7,"position_changes":1,"current_zones":["street_zone"],"entered_zones":["driveway_car_zone","street_zone"],"has_clip":false,"has_snapshot":true},"type":"update"}
{"before":{"id":"1670460861.454207-v45ue5","camera":"driveway_cam","frame_time":1670460875.598356,"snapshot_time":1670460862.198257,"label":"car","sub_label":null,"top_score":0.74609375,"false_positive":false,"start_time":1670460861.454207,"end_time":null,"score":0.71484375,"box":[107,30,405,139],"area":32482, "ratio":2.7339449541284404,"region":[0,0,604,604],"stationary":false,"motionless_count":7,"position_changes":1,"current_zones":["street_zone"],"entered_zones":["driveway_car_zone","street_zone"],"has_clip":false,"has_snapshot":true}, "after":{"id":"1670460861.454207-v45ue5","camera":"driveway_cam","frame_time":1670460876.057111,"snapshot_time":1670460862.198257,"label":"car","sub_label":null,"top_score":0.74609375,"false_positive":false,"start_time":1670460861.454207,"end_time":1670460881.098627,"score":0.53515625,"box":[57,32,287,142],  "area":25300,"ratio":2.090909090909091,"region":[0,0,668,668],"stationary":false,"motionless_count":1,"position_changes":1,"current_zones":["street_zone"],"entered_zones":["driveway_car_zone","street_zone"],"has_clip":false,"has_snapshot":true},"type":"end"}

FFprobe output from your camera

Not sure how to do this

Frigate stats

{"baby_cam":{"camera_fps":1.0,"capture_pid":230,"detection_fps":0.0,"pid":219,"process_fps":1.0,"skipped_fps":0.0},"detection_fps":0.0,"detectors":{"coral":{"detection_start":0.0,"inference_speed":10.09,"pid":216}},"doorbell_cam":{"camera_fps":5.0,"capture_pid":241,"detection_fps":0.0,"pid":229,"process_fps":5.0,"skipped_fps":0.0},"driveway_cam":{"camera_fps":5.1,"capture_pid":231,"detection_fps":0.0,"pid":220,"process_fps":5.1,"skipped_fps":0.0},"front_yard_cam":{"camera_fps":4.9,"capture_pid":235,"detection_fps":0.0,"pid":223,"process_fps":4.9,"skipped_fps":0.0},"service":{"latest_version":"0.11.1","storage":{"/dev/shm":{"free":4102.4,"mount_type":"tmpfs","total":4113.0,"used":10.6},"/media/frigate/clips":{"free":210672.3,"mount_type":"ext4","total":245373.3,"used":24689.9},"/media/frigate/recordings":{"free":210672.3,"mount_type":"ext4","total":245373.3,"used":24689.9},"/tmp/cache":{"free":4113.0,"mount_type":"tmpfs","total":4113.0,"used":0.0}},"temperatures":{},"uptime":90431,"version":"0.11.1-2eada21"}}

Operating system

HassOS

Install method

HassOS Addon

Coral version

USB

Network connection

Wired

Camera make and model

Dahua 6L04351PAGC8C7A

Any other information that may be helpful

No response

NickM-27 commented 1 year ago

I'd recommend watching the debug view of the camera (cameras -> open camera -> choose debug in top right -> enabled zones and bounding boxes)

This code hasn't changed in a long time and I've never seen an issue so it's likely a config error. Due be aware objects presence in a zone is based on the bottom center of its bounding box. Object masks work similarly as well.

We'll need a screen recording of the debug view along with MQTT output to be able to identify any issue.

J-Pipe commented 1 year ago

Well I think the first "bug" (unless it is supposed to work like this?) is that if the "type" is "new", then the current_zones seems to be always empty in the "before" message. Not sure if that is intended, but seems like a bug to me.

Below is the example picture. You can see the street_zone outlined in red. The yellow squares are from the 4th and 6th messages in the MQTT messages (linked below) but the "current_zones" is empty. I am also linking the exported video from that was used for this test.

MQTT_Messages.txt Debug Video Debug_Pic

NickM-27 commented 1 year ago

Thanks for the detailed video and messages. Upon looking at the messages it seems to be working as expected. You can notice that in the second message:

{
    "before": {
        "id": "1670559975.316493-6sfz4l",
        "camera": "driveway_cam",
        "frame_time": 1670559976.014847,
        "snapshot_time": 1670559976.014847,
        "label": "car",
        "sub_label": null,
        "top_score": 0.70703125,
        "false_positive": false,
        "start_time": 1670559975.316493,
        "end_time": null,
        "score": 0.7890625,
        "box": [
            346,
            322,
            995,
            719
        ],
        "area": 257653,
        "ratio": 1.6347607052896724,
        "region": [
            0,
            0,
            1296,
            1296
        ],
        "stationary": false,
        "motionless_count": 2,
        "position_changes": 1,
        "current_zones": [
            "driveway_car_zone"
        ],
        "entered_zones": [
            "driveway_car_zone"
        ],
        "has_clip": false,
        "has_snapshot": true
    },
    "after": {
        "id": "1670559975.316493-6sfz4l",
        "camera": "driveway_cam",
        "frame_time": 1670559977.219545,
        "snapshot_time": 1670559976.014847,
        "label": "car",
        "sub_label": null,
        "top_score": 0.7265625,
        "false_positive": false,
        "start_time": 1670559975.316493,
        "end_time": null,
        "score": 0.546875,
        "box": [
            616,
            210,
            758,
            326
        ],
        "area": 16472,
        "ratio": 1.2241379310344827,
        "region": [
            0,
            0,
            1464,
            1464
        ],
        "stationary": false,
        "motionless_count": 0,
        "position_changes": 1,
        "current_zones": [],
        "entered_zones": [
            "driveway_car_zone"
        ],
        "has_clip": false,
        "has_snapshot": true
    },
    "type": "update"
}

the before.box and after.box are very different, along with the before.area and after.area:

{
    "before": {
        "id": "1670559975.316493-6sfz4l",
        "box": [
            346,
            322,
            995,
            719
        ],
        "area": 257653,
    },
    "after": {
        "id": "1670559975.316493-6sfz4l",
        "box": [
            616,
            210,
            758,
            326
        ],
        "area": 16472,
    },
    "type": "update"
}

this can happen because frigate's tracking is somewhat simple. If it sees multiple of the same object it doesn't use things like size, color, direction of travel to differentiate between them across frames. This means that sometimes frigate will swap which car is which. In this case it appears that frigate thinks the car is in a different location for some of these frames which is why its current_zone is empty.

You can see this switch back in the third message:

{
    "before": {
        "id": "1670559975.316493-6sfz4l",
        "box": [
            616,
            210,
            758,
            326
        ],
        "area": 16472,
        "current_zones": [],
        "entered_zones": [
            "driveway_car_zone"
        ],
    },
    "after": {
        "id": "1670559975.316493-6sfz4l",
        "box": [
            494,
            62,
            798,
            483
        ],
        "area": 127984,
        "current_zones": [
            "driveway_car_zone"
        ],
        "entered_zones": [
            "driveway_car_zone"
        ],
    },
    "type": "update"
}
NickM-27 commented 1 year ago

Also as far as the type: new goes, it will not have before.current_zones because "false_positive": true, I believe that is working as expected as well.

J-Pipe commented 1 year ago

Sure, I understand that the tracking is not perfect, but for the fourth and sixth messages specifically, how can you explain that after.current_zones is empty, when the after.box is clearly in the street_zone?

NickM-27 commented 1 year ago

how can you explain that after.current_zones is empty, when the after.box is clearly in the street_zone?

this is the fourth message:

{
    "before": {
        "id": "1670559975.316493-6sfz4l",
        "camera": "driveway_cam",
        "frame_time": 1670559977.611997,
        "snapshot_time": 1670559976.014847,
        "label": "car",
        "sub_label": null,
        "top_score": 0.7265625,
        "false_positive": false,
        "start_time": 1670559975.316493,
        "end_time": null,
        "score": 0.609375,
        "box": [
            494,
            62,
            798,
            483
        ],
        "area": 127984,
        "ratio": 0.7220902612826603,
        "region": [
            196,
            0,
            804,
            608
        ],
        "stationary": false,
        "motionless_count": 0,
        "position_changes": 1,
        "current_zones": [
            "driveway_car_zone"
        ],
        "entered_zones": [
            "driveway_car_zone"
        ],
        "has_clip": false,
        "has_snapshot": true
    },
    "after": {
        "id": "1670559975.316493-6sfz4l",
        "camera": "driveway_cam",
        "frame_time": 1670559980.810574,
        "snapshot_time": 1670559976.014847,
        "label": "car",
        "sub_label": null,
        "top_score": 0.7265625,
        "false_positive": false,
        "start_time": 1670559975.316493,
        "end_time": null,
        "score": 0.53515625,
        "box": [
            90,
            71,
            441,
            249
        ],
        "area": 62478,
        "ratio": 1.9719101123595506,
        "region": [
            0,
            0,
            1312,
            1312
        ],
        "stationary": false,
        "motionless_count": 0,
        "position_changes": 1,
        "current_zones": [],
        "entered_zones": [
            "driveway_car_zone"
        ],
        "has_clip": false,
        "has_snapshot": true
    },
    "type": "update"
}

notice the "score": 0.53515625, which is below what you have set in your zone filter config

  driveway_cam:
    ...
    ...
    zones:
      # Required: name of the zone
      # NOTE: This must be different than any camera names, but can match with another zone on another
      #       camera.
      street_zone:          
        coordinates: 418,282,559,278,527,195,951,226,941,179,782,140,483,96,195,80,0,84,0,189,144,181,154,295,257,290
        objects:
          - car
          - person
        filters:
          car:
            min_area: 1800
            max_area: 755000
            min_score: 0.5
            threshold: 0.6 # <--- greater than the score in the mqtt message
J-Pipe commented 1 year ago

I mean, it seems like it should still say what zone it’s in. In the last message, the score is 59% and that’s when it finally picks up the street_zone 🤔

NickM-27 commented 1 year ago

I mean, it seems like it should still say what zone it’s in.

Your config is specifically telling frigate not to consider an object with a score below 60% as part of the zone, so that would be entirely incorrect if it was always listed in current_zones despite that.

Also, by default frigate does not filter zone presence by score so I'd recommend removing that filter entirely for all zones.

I ran your debug clip multiple times and you can see the different scores having a different result in the zone

Screen Shot 2022-12-09 at 08 56 09 AM

NickM-27 commented 1 year ago

I removed the filters from the zone and it now is working consistently

Screen Shot 2022-12-09 at 09 13 03 AM

J-Pipe commented 1 year ago

Thank you Nick. I’ll try to play with that tonight and see what results I get.

J-Pipe commented 1 year ago

So, why would before:false_positive always be true for type:new objects?

J-Pipe commented 1 year ago

So, why would before:false_positive always be true for type:new objects?

Nevermind, found the answer: "Message published for each changed event. The first message is published when the tracked object is no longer marked as a false_positive."

So my thinking is that it should still record the current_zone even if it's a false positive. One issue that I'm having is that my car is in the street zone in the before message, but in the driveway zone in the after. Since the MQTT message is not including the proper zone when false_positive is true, it throws things off.

NickM-27 commented 1 year ago

Since the MQTT message is not including the proper zone when false_positive is true, it throws things off.

This is likely due to fps being 1, if the fps is increased even to 2 it would likely be improved.

J-Pipe commented 1 year ago

fps is 5 for all my cameras except baby_cam. I am past the point of trying to debug my situation and this is about the functionality of Frigate. I think the current_zone (and probably entered_zones) should be populated in the first MQTT message of the event. When that message is sent, we know it’s not a false_positive anyways (as that is what triggers the first message to transmit)

NickM-27 commented 1 year ago

fps is 5 for all my cameras except baby_cam

my bad, looked at the wrong one, in any case the idea is the same that increasing the fps is the recommendation.

I have a similar setup with a driveway_entrance and driveway zone, only alerting when a car was in driveway_entrance and is now in driveway, and with 8 fps set I have not missed a notification with that setup.

J-Pipe commented 1 year ago

I can try increasing it but I am past the point of trying to debug my situation and this is about the current functionality of Frigate. I think the current_zone (and probably entered_zones) should be populated in the first MQTT message of the event. When that message is sent, we know it’s not a false_positive anyways (as that is what triggers the first message to transmit). I can try to fix it and submit a PR if you think it will get accepted.

NickM-27 commented 1 year ago

I think the current_zone (and probably entered_zones) should be populated in the first MQTT message of the event.

I mean it is populated in the first message, just after because before the object was a true positive, the object was a false positive.

When an object is a false positive frigate is not confident that the object is what the label says. This means that object can not be part of a zone, among other things. The zone not being included in the before of a new event is not because it just isn't included, the object does not have any zones attached since a false positive doesn't have presence in zones.

This means the fix is not as simple as including information that exists but isn't presented, it'll need to make large changes to either fundamentally change the way frigate treats false positives & zones or have it keep that information just for mqtt in a separate list. It will also be a large breaking change for all existing automations that work off the idea that an object being in a zone means it is not a false positive.

To me, this is an example of https://xyproblem.info/ where the mqtt information is being treated as the problem but in actuality the problem is that frigate does not have enough time to be confident that the car is actually a car before it enters the driveway. https://docs.frigate.video/guides/false_positives also explains how this works.

I can try to fix it and submit a PR if you think it will get accepted.

As you can probably tell, I disagree with this. However, I'm always open to hearing other's opinions / perspectives. Also, what is accepted is up to @blakeblackshear and not me 👍

J-Pipe commented 1 year ago

The issue of current_zones not being populated (even on the type:update objects) continues to be an issue for me that I cannot resolve. Here is an interesting case that happened today...

In the first after object, the car should be in the driveway_car_zone, but instead, current_zones is empty. Looking at the other data, it's not a false_positive, and the score/area/ratio all meet my filters. ODDLY, in the very next MQTT message, the before object does include the proper current_zones. What would cause this?? I can provide more data if needed (video, screenshots, config info, camera/zone debug info).

{"before":{"id":"1672102387.274661-4gigs3","camera":"driveway_cam","frame_time":1672103572.98488,"snapshot_time":1672103572.523118,"label":"car","sub_label":null,"top_score":0.84375,"false_positive":false,"start_time":1672102387.274661,"end_time":null,"score":0.71484375,"box":[367,111,503,256],"area":19720,"ratio":0.9379310344827586,"region":[175,0,535,360],"stationary":false,"motionless_count":5,"position_changes":3,"current_zones":["street_zone"],"entered_zones":["street_zone"],"has_clip":false,"has_snapshot":false},"after":{"id":"1672102387.274661-4gigs3","camera":"driveway_cam","frame_time":1672103574.78718,"snapshot_time":1672103574.78718,"label":"car","sub_label":null,"top_score":0.84375,"false_positive":false,"start_time":1672102387.274661,"end_time":null,"score":0.7265625,"box":[350,132,539,317],"area":34965,"ratio":1.0216216216216216,"region":[251,34,635,418],"stationary":false,"motionless_count":2,"position_changes":3,"current_zones":[],"entered_zones":["street_zone"],"has_clip":false,"has_snapshot":false},"type":"update"}}
{"before":{"id":"1672102387.274661-4gigs3","camera":"driveway_cam","frame_time":1672103574.78718,"snapshot_time":1672103574.78718,"label":"car","sub_label":null,"top_score":0.84375,"false_positive":false,"start_time":1672102387.274661,"end_time":null,"score":0.7265625,"box":[350,132,539,317],"area":34965,"ratio":1.0216216216216216,"region":[251,34,635,418],"stationary":false,"motionless_count":2,"position_changes":3,"current_zones":["driveway_car_zone"],"entered_zones":["street_zone","driveway_car_zone"],"has_clip":true,"has_snapshot":true},"after":{"id":"1672102387.274661-4gigs3","camera":"driveway_cam","frame_time":1672103580.191988,"snapshot_time":1672103580.191988,"label":"car","sub_label":null,"top_score":0.84375,"false_positive":false,"start_time":1672102387.274661,"end_time":null,"score":0.83203125,"box":[371,199,705,501],"area":100868,"ratio":1.1059602649006623,"region":[122,0,1250,1128],"stationary":false,"motionless_count":13,"position_changes":3,"current_zones":["driveway_car_zone"],"entered_zones":["street_zone","driveway_car_zone"],"has_clip":true,"has_snapshot":true},"type":"update"}}
github-actions[bot] commented 1 year ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

blakeblackshear commented 1 year ago

Looks like over 10 frames were processed in between these two mqtt messages. A message isn't sent for each frame.

github-actions[bot] commented 1 year ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.