[Config Support]: Hardware Acel, high spiking CPU usage on detection

[storxusmc]

Describe the problem you are having

I am trying to figure out if my config is poorly setup or if I'm obviously missing something. I have been experimenting with frigate with my reolink doorbell with the intentions of adding possibly up to 5 cameras once I get everything figured out, but I keep having issues with false positives of people detected and CPU usage on my frigate computer spiking to 100% usage when someone walks in view of the camera and it detects motion. Example last night around 8:13pm, it triggered that person was detected 51 times in a row, each time there is no visible person in the video. I have attempted to limit this with a detection zone that has no plants, but no luck. I reached out on Reddit and they told me i need to figure out my hardware accel for the CPU usage, but I'm not sure, I tried adding the CPU, but with both the Coral and CPU enabled it seemed to increase the average load on the CPU.

Version

SYSTEM 0.13.2-6476F8A

Frigate config file

mqtt:
  host: 192.168.4.69
  user: mqtt-user
  password: <REMOVED>

#Global Object Settings
objects:
  track:
    - person
  filters:
    person:
      min_area: 5000

detectors:
  coral:
    type: edgetpu
    device: usb

# if you want to decode a h264 stream
#  ffmpeg:
#    hwaccel_args: preset-intel-qsv-h264

cameras:
  Front_Door:
    ffmpeg:        
      inputs:
        - path: rtsp://camera:<REMOVED>@192.168.4.117:554/h264Preview_01_main
          roles:
            - record
            - detect
        - path: rtsp://camera:<REMOVED>@192.168.4.117:554/h264Preview_01_sub
          roles:
            - rtmp
    detect:
      width: 1280
      height: 720

#    record:
#      width: 2560
#      height: 1920

    snapshots:
      enabled: true
      
    record: 
      enabled: true
      retain: 
        days: 21
      events:
        retain: 
          default: 42

    zones:
      Front_Door_Zone:
        coordinates: 0,1920,1388,1920,1654,1920,1750,1487,1781,1040,1765,512,832,543,751,1083,320,1112,0,1121
        objects: 
          - person
          #- cat
        #filters: 
          #person:
            #min_area: 5000
            #max_area: 100000

Relevant log output

2024-07-04 12:53:18.587732198  [INFO] Preparing Frigate...
2024-07-04 12:53:18.598382016  [INFO] Starting Frigate...
2024-07-04 12:53:19.666704388  [2024-07-04 12:53:19] frigate.app                    INFO    : Starting Frigate (0.13.2-6476f8a)
2024-07-04 12:53:19.693698431  [2024-07-04 12:53:19] peewee_migrate.logs            INFO    : Starting migrations
2024-07-04 12:53:19.696931083  [2024-07-04 12:53:19] peewee_migrate.logs            INFO    : There is nothing to migrate
2024-07-04 12:53:19.701208248  [2024-07-04 12:53:19] frigate.app                    INFO    : Recording process started: 435
2024-07-04 12:53:19.703646037  [2024-07-04 12:53:19] frigate.app                    INFO    : go2rtc process pid: 89
2024-07-04 12:53:19.724473096  [2024-07-04 12:53:19] frigate.app                    INFO    : Output process started: 447
2024-07-04 12:53:19.738606881  [2024-07-04 12:53:19] frigate.app                    INFO    : Camera processor started for Front_Door: 455
2024-07-04 12:53:19.745901239  [2024-07-04 12:53:19] frigate.app                    INFO    : Capture process started for Front_Door: 457
2024-07-04 12:53:22.864064982  [2024-07-04 12:53:19] detector.coral                 INFO    : Starting detection process: 445
2024-07-04 12:53:22.866345332  [2024-07-04 12:53:19] frigate.detectors.plugins.edgetpu_tfl INFO    : Attempting to load TPU as usb
2024-07-04 12:53:22.866427423  [2024-07-04 12:53:22] frigate.detectors.plugins.edgetpu_tfl INFO    : TPU found

Frigate stats

No response

Operating system

UNRAID

Install method

Docker Compose

Object Detector

Coral

Any other information that may be helpful

Intel® Core™ i3-10100 CPU
ASRock H470M-ITX/ac
Memory: 32 GiB DDR4
Google Coral TPU

[hawkeye217]

You'll want to enable hardware acceleration. See the docs: https://docs.frigate.video/configuration/hardware_acceleration

I'd recommend using preset-vaapi for your hwaccel args.

And you'll also want to set up the doorbell with the recommended Reolink config. See those docs here:
https://docs.frigate.video/configuration/camera_specific/#reolink-cameras
https://docs.frigate.video/configuration/camera_specific/#reolink-doorbell

[storxusmc]

So i added the hwaccell args to the config, but it brought my overall cpu usage on the server up slightly and reduce the usage on the coral tpu... is this normal?

Question:
Does the go2rtc lines replace the current camera section in my config? or is this in addition on the config?

It did populate the GPU section on the system.

[NickM-27]

what is your CPU? 18 fps also seems like it may be detecting a lot of motion

[storxusmc]

Intel® Core™ i3-10100 CPU
ASRock H470M-ITX/ac
Memory: 32 GiB DDR4
Google Coral TPU

[hawkeye217]

You'll want to watch the debug view to see what's happening on the camera. If you see lots of red boxes, you'll want to tune your motion settings. See the docs:

https://docs.frigate.video/configuration/motion_detection