[Support]: Detection Behavior

[percy-punch]

Describe the problem you are having

HI,
New to frigate and it has a big learning curve!

I'm using OpenVino for object detection using gpu pass through and i have a low inference <30ms on average it. I have an issue with my front camera in that i was getting some false positives so i started adding masks and zones to reduce them and also added filters: This worked well but now i'm struggling to detect a person consistently. It seems to work best closer to my front door but as you move away it's loses tracking but maintains tracking boxes which do remain above the threshold i have set of 2500.
The other cameras with basic configs identify consistently but they are in a less challenging location.

This is obviously a problem with my config and lack of knowledge i would be grateful for any feedback so i can correct this issues.

Version

0.13.2

Frigate config file

mqtt:
  host: 192.168.2.201 # <---- same mqtt broker that home assistant uses
  user: mqttuser
  password: ******
  port: 1883
ffmpeg:
  hwaccel_args: preset-vaapi

detectors:
  ov:
    type: openvino
    device: AUTO
    model:
      path: /openvino-model/ssdlite_mobilenet_v2.xml

model:
  width: 300
  height: 300
  input_tensor: nhwc
  input_pixel_format: bgr
  labelmap_path: /openvino-model/coco_91cl_bkgr.txt
#Global Object Settings
objects:
  track:
  - person
  filters:
    person:
      min_area: 2500
      max_area: 20000
#Global Record settings
record:
  enabled: true
  retain:
    days: 3
    mode: all
  events:
    retain:
      default: 30
      mode: motion
#rtpc stream setup
go2rtc:
  streams:
    FrontCam:
        #"ffmpeg: " gives smooth video in mse stream
    - ffmpeg:rtsp://admin:***@192.168.1.51:554/cam/realmonitor?channel=1&subtype=0
    RearCam:
    - ffmpeg:rtsp://admin:***@192.168.1.76:554/cam/realmonitor?channel=1&subtype=0
    SideCam:
    - ffmpeg:rtsp://admin:***@192.168.1.52:554/cam/realmonitor?channel=1&subtype=0

#Camera Setup
cameras:
  FrontCam:
    ffmpeg:
      inputs:
        # High Resolution Stream
      - path: rtsp://admin:***@192.168.1.51:554/cam/realmonitor?channel=1&subtype=0
        input_args: preset-rtsp-restream
        roles:
        - record
        # Low Resolution Stream
      - path: rtsp://admin:***@192.168.1.51:554/cam/realmonitor?channel=1&subtype=1
        input_args: preset-rtsp-restream
        roles:
        - detect
    detect:
      width: 640
      height: 480
      fps: 5
    objects:
      track:
        - person
        #- car
        - bicycle
      filters:
        person:
          min_area: 2500
          max_area: 20000
          threshold: 0.7
    record:
      events: 
        required_zones:
        - zone_front
    motion:
      mask:
      - 640,0,590,96,397,34,412,0
      - 215,480,326,354,156,288,72,119,0,116,0,480
      - 255,333,292,192,211,219
      - 0,0,54,0,91,0,223,0,319,0,182,33,0,80
    
    zones:
      zone_front:
        coordinates: 
          307,47,322,59,344,62,356,51,398,42,436,64,464,78,503,107,517,121,545,141,567,162,562,188,549,233,529,276,465,364,392,445,323,390,290,272,356,218,315,156,268,102,234,60
        
  RearCam:
    ffmpeg:
      inputs:
        # High Resolution Stream
      - path: rtsp://admin:***@192.168.1.76:554/cam/realmonitor?channel=1&subtype=0
        input_args: preset-rtsp-restream
        roles:
        - record
        # Low Resolution Stream
      - path: rtsp://admin:***@192.168.1.76:554/cam/realmonitor?channel=1&subtype=1
        input_args: preset-rtsp-restream
        roles:
        - detect
    detect:
      width: 640
      height: 480
      fps: 5
    motion:
      mask:
      - 0,0,121,0,54,480,0,480
      - 640,186,640,0,129,0,105,83
      - 640,480,561,480,564,390,640,249
  SideCam:
    ffmpeg:
      inputs:
        # High Resolution Stream
      - path: rtsp://admin:***@192.168.1.52:554/cam/realmonitor?channel=1&subtype=0
        input_args: preset-rtsp-restream
        roles:
        - record
        # Low Resolution Stream
      - path: rtsp://admin:***@192.168.1.52:554/cam/realmonitor?channel=1&subtype=1
        input_args: preset-rtsp-restream
        roles:
        - detect
    detect:
      width: 640
      height: 480
      fps: 5

    motion:
      mask:
      - 497,480,640,480,640,0,262,0,229,48
      - 240,30,67,28,67,173,75,320,118,480,0,480,0,0,240,0

Relevant log output

2024-06-04 15:13:39.709835215  [INFO] Preparing Frigate...
2024-06-04 15:13:39.768308698  [INFO] Starting Frigate...
2024-06-04 15:13:41.520744331  [2024-06-04 15:13:41] frigate.app                    INFO    : Starting Frigate (0.13.2-6476f8a)
2024-06-04 15:13:41.642158218  [2024-06-04 15:13:41] frigate.config                 WARNING : Customizing more than a detector model path is unsupported.
2024-06-04 15:13:41.663863709  [2024-06-04 15:13:41] peewee_migrate.logs            INFO    : Starting migrations
2024-06-04 15:13:41.672153846  [2024-06-04 15:13:41] peewee_migrate.logs            INFO    : There is nothing to migrate
2024-06-04 15:13:41.678783715  [2024-06-04 15:13:41] frigate.app                    INFO    : Recording process started: 290
2024-06-04 15:13:41.679828378  [2024-06-04 15:13:41] frigate.app                    INFO    : go2rtc process pid: 89
2024-06-04 15:13:41.705705995  [2024-06-04 15:13:41] frigate.app                    INFO    : Output process started: 301
2024-06-04 15:13:41.732589188  [2024-06-04 15:13:41] detector.ov                    INFO    : Starting detection process: 299
2024-06-04 15:13:42.097425805  [2024-06-04 15:13:42] frigate.app                    INFO    : Camera processor started for FrontCam: 325
2024-06-04 15:13:42.102653053  [2024-06-04 15:13:42] frigate.app                    INFO    : Camera processor started for RearCam: 326
2024-06-04 15:13:42.112348542  [2024-06-04 15:13:42] frigate.app                    INFO    : Camera processor started for SideCam: 327
2024-06-04 15:13:42.132841524  [2024-06-04 15:13:42] frigate.app                    INFO    : Capture process started for FrontCam: 331
2024-06-04 15:13:42.143215825  [2024-06-04 15:13:42] frigate.app                    INFO    : Capture process started for RearCam: 334
2024-06-04 15:13:42.143220141  [2024-06-04 15:13:42] frigate.app                    INFO    : Capture process started for SideCam: 338
2024-06-04 15:13:42.331917002  �[35mE: [xLinkUsb] [    422331] [frigate.detecto] usb_find_device_with_bcd:266	Library has not been initialized when loaded�[0m
2024-06-04 15:13:42.340184462  �[35mE: [xLinkUsb] [    422340] [frigate.detecto] usb_find_device_with_bcd:266	Library has not been initialized when loaded�[0m
2024-06-04 15:13:42.732243041  [2024-06-04 15:13:42] frigate.detectors.plugins.openvino INFO    : Model Input Shape: [1,300,300,3]
2024-06-04 15:13:42.732847772  [2024-06-04 15:13:42] frigate.detectors.plugins.openvino INFO    : Model Output-0 Shape: [1,1,100,7]
2024-06-04 15:13:42.733215354  [2024-06-04 15:13:42] frigate.detectors.plugins.openvino INFO    : Model has 1 Output Tensors

FFprobe output from your camera

FFPROBE OUTPUT
Stream 0:
Return Code: 0

Video:

Codec: H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10
Resolution: 2592x1944
FPS: 15/1

Audio:

Codec: AAC (Advanced Audio Coding)

Stream 1:
Return Code: 0

Video:

Codec: H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10
Resolution: 640x480
FPS: 5/1

Frigate stats

No response

Operating system

Proxmox

Install method

Docker Compose

Coral version

Other

Network connection

Wired

Camera make and model

Dahua

Any other information that may be helpful

No response

[NickM-27]

if I had to guess your min_area is too high, causing the person to get filtered out

[percy-punch]

That kind of what i thought but the bunding box when selected gives you a pixel count and its above the threshold.

[NickM-27]

not sure what you mean, there is no bounding box when the person is further away, just a motion box

[percy-punch]

sorry yes motion box.

I think you are correct actually i have been playing with the settings lowering min_area:750 seems to help. I also think the threshold: 0.7should be lowered a touch as well. My assumption is a blue motion box changing from yellow to blue means the object has gone below the threshold? hard to see on a mobile screen the black text on blue.
I think the fundamental issue is my camera is high up with a large field of view. Detections will be high and persons far from the house will have fewer pixels on the screen so i need to find a balance.