Local realtime person detection for RTSP cameras

I think the issue was either the docker build needed updating, the coral drivers or both and a hardware reboot of my unRAID server. TBH I can’t remember. Sorry

I seem to not be retaining the pre_capture and post_capture periods for my cameras.
I have a significant portion of one camera masked (busy street) so that I only get an event when something moves on my property. When something does get detected on my property i’d still like to see what the preceding 15 seconds looked like. I thought the below would achieve this but my events and recordings seem to have only a few seconds before the property boundary and some events are less than 10 seconds of playback in total (i assumed it would be minimum of 25 with these settings).

record:
  enabled: true
  retain:
    days: 1
    mode: motion
  events:
    max_seconds: 1800
    pre_capture: 15
    post_capture: 10
    objects:
      - person
    required_zones: []
    retain:
      mode: active_objects
      default: 5
      objects:
        person: 7
        dog: 2

Retain mode active_objects will only ever retain segments with actively moving confirmed objects. If your pre/post capture segments don’t contain active objects, they will be discarded regardless of the setting.

Maybe you already have, but requied zones are generally the recommended approach to only getting events on your property, not masks.

Please create a github issue if you want further support.

I folks, I saw that 2days ago version 11 was released,

but I don’t see the aarch64, I need to wait or aarch64 is not longer available?

Did you figure this out?
I’m seeing this as well.
thanks

Lovely, easy. that means I can use it for raspberry, @AngellusMortis correct?

How should I select it with the tag?

blakeblackshear/frigate: stable ???

Hi,

Reported before, still trying to figure out what can cause it. Not always but happens frequently, I get continues update of an event. The update is sent every minute until I restart frigate or a new object is detected. You can see from the screen shoot the continues events with MQTT sniffer.

Any idea what is the cause?

I’m trying to take advantage of the inbuilt nvidia decoding with version 0.11 but having some issues, wonder if anyone has any ideas?

NUC 11 enthusiasts running HA OS (no virtualisation)
Frigate installed on the same box as an add-on
RTX2060 graphics

When running:
ffmpeg -decoders | grep cuvid
I get the expected response showing me the available codec.

But when I then run:
nvidia-smi
I get no such file

Frigate does start but none of the feeds are working, green screens.

Any suggestions?

Help With config please.
After update to 0.11.0 some changes were broken, please help non of the cameras are working.
HAOS with Frigate full access addon
Previously working config with stimeout changed:

ffmpeg:

  global_args: -hide_banner -loglevel warning

  hwaccel_args:
    - -hwaccel
    - qsv
    - -qsv_device
    - /dev/dri/renderD128

  input_args: -avoid_negative_ts make_zero -fflags +genpts+discardcorrupt -rtsp_transport tcp -timeout 5000000 -use_wallclock_as_timestamps 1

  output_args:

    detect: -f rawvideo -pix_fmt yuv420p

    record: -f segment -segment_time 10 -segment_format mp4 -reset_timestamps 1 -strftime 1 -c copy -an
    # Optional: output args for clips streams (default: shown below)
    #clips: -f segment -segment_time 10 -segment_format mp4 -reset_timestamps 1 -strftime 1 -c copy -an

Error now:

022-10-02 13:22:04] frigate.app                    INFO    : Capture process started for CAM5: 268
[2022-10-02 13:22:04] frigate.app                    INFO    : Capture process started for CAM6: 277
[2022-10-02 13:22:04] frigate.app                    INFO    : Capture process started for CAM7: 280
[2022-10-02 13:22:04] frigate.app                    INFO    : Capture process started for CAM8: 281
[2022-10-02 13:22:04] frigate.app                    INFO    : Capture process started for CAM_UA_HOME_1: 286
[2022-10-02 13:22:04] frigate.app                    INFO    : Capture process started for CAM_UA_HOME_2: 294
[2022-10-02 13:22:04] frigate.edgetpu                INFO    : TPU found
[2022-10-02 13:22:04] frigate.app                    INFO    : Capture process started for CAM_UA_HOME_3: 310
[2022-10-02 13:22:04] frigate.app                    INFO    : Capture process started for CAM_UA_HOME_4: 319
[2022-10-02 13:22:04] ws4py                          INFO    : Using epoll
[2022-10-02 13:22:05] frigate.video                  ERROR   : CAM_technicka: Unable to read frames from ffmpeg process.
[2022-10-02 13:22:05] frigate.video                  ERROR   : CAM_technicka: ffmpeg process is not running. exiting capture thread...
[2022-10-02 13:22:06] frigate.video                  ERROR   : CAM8: Unable to read frames from ffmpeg process.
[2022-10-02 13:22:06] frigate.video                  ERROR   : CAM8: ffmpeg process is not running. exiting capture 

When i removed hwaccel_args

hwaccel_args: []

It started to work, but i am confused as if its important or not oO
I have HAOS in a VM. Please advise

detectors:
  coral_pci:
    type: edgetpu
    device: pci

Dramatically increased RAM usage since upgrade. I have 32gb RAM installed. Frigate now uses nearly all of it:

It took off after updating on Sep 25:

Is this the expected behavior? Thanks

1 Like

Obviously not expected. Please create an issue on github. We don’t provide support in this thread.

1 Like

Apologies if this has already been asked and answered as it’s a very long thread. Can this be used to recognize specific people, or is this purely person vs smth else categorization?

no, I’m still trying to figure it out

HI anyone can help me to understand how to reference the new 11 release?
i see this tag
docker pull blakeblackshear/frigate: stable

but until today I used this one
mage: blakeblackshear/frigate:stable-aarch64

How can I reference the arm64? My hw is a raspberry pi 4

regards

I think its possible but its hard or use Frigate+

Look at dockerhub under “tags”. Look like all OS are now combined under same image

https://hub.docker.com/r/blakeblackshear/frigate/tags?page=2

docker pull blakeblackshear/frigate:stable

docker pull blakeblackshear/frigate:0.11.0

EDIT

From changelog

Updated Dockerbuild
Frigate is now built on top of plain Debian containers and consolidated into a single multiarch image. Support for nvidia decoding is built into the single image. There is no longer a need for a special tag.

2 Likes

Thanks for this thread. I was quite frustrated until I found this.

Just to add: I also created a dedicated user for Frigate (within the Doorbird app). The ONLY right I assigned to this user was the first option “Live Stream”. (Sorry, wording might not fit, my Doorbird app is German language saying “Immer live sehen”). With this user and single right it worked like a charm!

Before I had multiple other options set, and Frigate could not read the stream…

HI All. I’m a bit stumped .
I have a intel nuc running HAOS (latest ver 2002.10 ,operating system 8.5). I also have a google coral usb but cannot get it to be detected in frigate addon. I have run docker on the same machine before and frigate and the coral is detected. is there a driver i need to install to get it working with home assistant OS?

did you ever figure this out? also couldnt find the issue if you made one. i am also seeing memory issues.