Local realtime person detection for RTSP cameras

I’m not even expecting help from you. Nothing constructive? Better to stay silent…

What version frigate?
What system OS?
How installed?
What camera using?

Looking at log it look like immediate crash but not sure. Can you verify log above is after some hours use?

I using several months now with never freezing or need to restart due to software error. In all case in this forum I think user, hardware, OS or camera specific issue and mostly at time of install or hardware change.

When I think…could you be running out of memory or storage? I had problem in past month where I run out of storage and cause similar error in log. I reduce retention days and solved. I think clips are in memory so I guess possible low memory create similar condition. Are you using high res image for clips? (Posting config and you easy we know this answer) maybe check using with low res stream for clips and hi res for recording if not already doing.

Please can you show a simple way to install frigate in a docker , for a noob ?

I didn’t find a tutorial for that.

I wish to use my nuc with gpu passthrough and my usb google coral, with frigate in HA.

For now i use the entire machine and works, but i need to use proxmox to get other vm for other services.

Have you seen this documentation?

https://blakeblackshear.github.io/frigate/installation

i used docker-compose and it is really easy.
you only need to redefine your path for persistency:

  - <path_to_config_file>:/config/config.yml:ro
  - <path_to_directory_for_media>:/media/frigate

I’m using Frigate with a Beelink U55 with 3 cameras and it works very very well. No crash and no freeze for weeks

Cpu always under 25% and no memory issues. I will improve my cmaeras numbers
It’s a dedicated mini-pc for NVR

What hardware do you have?

thanks but in proxmox how can i create it?

i didn’t found a step by step :frowning:

Hello, did you get to record audio?

@Hachee https://blakeblackshear.github.io/frigate/troubleshooting#how-can-i-get-sound-or-audio-in-my-clips-and-recordings

Having some issues with high cpu usage…
Running 8 cams where 5 of them is High-Res (2560x1920).
My CPU spikes at most 70% with an average of 50-60%.
Is this normal or is my ffmpeg settings all wrong?

What ffmpeg options would you use with an intel i5-6500T (Gen 5?)
I’ve set my ffmpeg as below, they sure are wrong…
Any advices are appreciated

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

Yes i see a lot of times, but in proxmox how can i create a lxc?

Step by step, here in this docs, is too generic.

In proxmox Create CT, then i select hardware and ubuntu about software, then i have ubuntu in docker, and now which command i need to launch, to create frigate in lxc?

similar here when i used proxmox and google coral.

you need to use in the host. I see that HA and frigate in vm on proxmox and google coral, with or without these settings about ffmpeg, same result, high cpu, but google coral used. And i get passthrough of GPU for HA vm.

For now i use HA in a nuc, baremetal, without vm and cpu is low.

So need to understand why in proxmox is too high.

ANd i need also understand setp by step how to create a lxc with frigate…

Not using Proxmox or VM … Baremetal…
It’s an HP mini 400 G2 ProDesk with an i5-6500T and 8gigs of ram, running Ubuntu 20 LTE.
What is your avg CPU usage?

Hi! I’m using the latest one with Coral.
Home Assistant on Ubuntu, Docker
Frigate is on Docker, Integration with HA via HACS custom component.
Axis M2026LE-MKII- if I’m using it’s native resolution, I get:

frigate    | ffmpeg.cam2.detect           ERROR   : rtsp://192.168.0.7:554/axis-media/media.amp?streamprofile=frigate: Server returned 5XX Server Error reply
frigate    | frigate.video                  INFO    : cam2: ffmpeg sent a broken frame. memoryview assignment: lvalue and rvalue have different structures

Whenever I use lower resolution, everything works well. I need to use as high resolution as possible cause the camera is installed on a decent height and people are rather small (need to be accurately detected), thus as it was recommended the optimal size of the object should be 300x300 or more.
With that in mind, I tried using the maximum resolution.
What I did to improve the situation - increased SHM SIZE to 2 GB.
Using 4 cameras totally now.

Thanks in advance for valuable advice!

Resolution may be specific to role. you can Record Hi res and Detect using Low res. Without showing your config or your detailed communcation of this information. It is very unclear what you are doing.

I think it mean resolution you have in config does not match resolution sent by camera

Detect on High res caused issues for me when doing this with (x8) camera, server would freeze and need reboot. I presume it was memory issue but never care to check further and just use low res since it worked more than adequately. Maybe only one camera need the detection with High Res. Maybe 1 camera high res detect and other low res. Freeze issue may also be a function of frame rate I guess. Again, without config it is impossible to know how you using. Maybe a lower frame rate will help

2 Likes

Hi all.
Just set up Frigate in docker on my Intel pc. Now I am thinking of buying a coral. What version/form factor is advisable to purchase? Probably the usb one would be more universal (as it could be usable with PI as well). Are there some drawbacks using usb coral in setup with ubuntu + Frigate docker containerm than the M2 versions ?

USB was slower but still pretty fast for me.
I had usb power issue that made PCIE better as well, USB would not connect after some reboots for reasons outside coral.

EDIT
Pricing may be a factor as well
I check today and USB is about double price of m.2 and PCIe. but they not so expensive so maybe not big deal
M.2 version now has DUAL TPU so this is very interesting to me!! Even with single TPU my current limit is due to Server CPU not able to process high res fast enough to feed TPU so even Dual may not have real benefit for me.

1 Like

@Minglarn Hi!

I was at my summer cottage and I tried to connect my 510WA with rtmp stream with VLC to be sure that configuration is correct before adding it to Frigate. Sorry to bother but if you add your rtmp address to vlc - do you see stream?

No worries.
I see the RTMP stream nice and crisp.

I have a zone shaping my fence.
Whenever somebody’s walking or driving close to the fence, the Motion Box overlaps my fence boundary and the detected object (beyon Zone) gets automatically associated with my Zone (unwanted behavior as I need to classify objects only within a Zone). How to avoid it?