Local realtime person detection for RTSP cameras

[scote (Scott Pattinson)]

I am interested in this too, I guess its a local addon to provide ingress access to frigate in hass?

[nwetp (Neil)]

@jasebob

This might be useful to others. Its an IOS notification to the HA app, where the notification action opens the application to frigate. It uses the single mode and a delay to prevent rapid retriggers which is simpler to get working than a templat condition. I wasn’t able to get notifications with embedded working when remote from my home network, but this is a good workaround. It would be nice if I could give the URL the event ID too, but the events page does not seem to be URL addressable.

This is really useful for me, I’ve got it working and prevents the repeated notifications, thanks for this.

The only thing I can’t seem to get working here (or using any of the other code from the docs) is to include the image in the notification for iOS.

Anyone got any ideas on how to add it?

[thundergreen]

well people… you can add an iframe to integrate frigate if you run frigate not as addon. If you run the addon iyou can add it to the menu. Who needs help to setup frigate via docker to iintegrate it in home assistant ? PM me fellows

[jasebob]

Yes, Raspberry Pi 4
It runs both HA and Frigate. Home Assistant via HAOS, and I just installed Frigate via the supervisor Add-on (and the Mosquitto MQTT broker installed that way too). I don’t have a Coral (although I’d like to get one when they come back into stock).
I have not done anything fancy with Docker, it just worked. But I am running aarch64-addon-frigate:1.13
The hardest part was getting an initial frigate.yml file going and figuring out my camera streams. Using VLC to connect to streams helped me to verify that I had the right camera information.

[jasebob]

I can’t get the included image working either.
This explains how to do it

Integration with Home Assistant | Frigate

But I cannot load this path via the nabu casa remote UI address.
https://HA_URL/api/frigate/notifications//thumbnail.jpg
Even creating the URL by hand from a past event I cannot get this working and I am wondering if it works within HAOS or if I need some docker arguments that are not thereby default to open up a port.

I could get all these HTTP APIs working from within my local network through

HTTP API | Frigate

So I was able to get a photo in the notification using a local address and it works if I am home and connected to my own wireless.
But my house also has windows so there isn’t a lot of point!

I would also note that I set the notification up to work on trigger.payload_json[’‘type’’] == ‘‘new’’ so it is fast and I find out as soon as a person enters the gate. Good for when I am home, but not as good to find the best image. If you use type == end, you will get a better image but the person may be gone before you get the notification.

[bremby]

@jasebob, thank you

Got it. I would like to split HA and Frigate
I configured everything on an Ubuntu VM, now I’m wondering if I can run docker frigate directly on the raspi without HAOS, I would like to know if someone already did it

Regards

[iMiMx (iMx)]

Managed to ditch Synology Surveillance Station (hurrah) and had fun playing with the various settings over the last few days, my USB Coral got delivered yesterday and wow what a difference that makes.

I have a few questions:

Zones, what consistutes in or out of a zone? Is it again the bottom centre of the bounding box? What happens if zones partially overlap?

In the documentation, I saw the below:

During testing, draw_zones should be set in the config to draw the zone on the frames so you can adjust as needed. The zone line will increase in thickness when any object enters the zone.

Does this still exist? I couldn’t see any reference in the wider documentation still, nor in the examples.

Also ‘min_score’ is not an option in a zone, this has to apply to the wider camera/global configuration?

P.S. I think this all really needs it’s own forum or sub-forum, in the 4000 above posts I’m sure there is all sorts of valuable information… but it’s a lot to read, even with search.

[hawkeye217]

@iMiMx

P.S. I think this all really needs it’s own forum or sub-forum, in the 4000 above posts I’m sure there is all sorts of valuable information… but it’s a lot to read, even with search.

The Discussions area on the Frigate Github has made conversations easier. Check it out here

[AussieByrd (Ben)]

Well, that was fast! Coral M.2 A+E arrived today - installed, drivers done, detectors changed across, Frigate running - success. Very easy setup; thing that took the longest was finding an m2 screw to mount the chip on my motherboard.

Running a CPU detector had my CPU ‘idle’ (normally ~1-10%) jump to ~25% at night (less movement I guess) to between 40% and 70% during the day. With the Coral installed, I’m pretty solid at ~20% day or night. So running Frigate is still a measurable hit on my CPU, but nothing like when I was running the CPU detector.

[AussieByrd (Ben)]

Hi @bremby… dunno about on a raspi, but that’s kinda what I’m doing on a repurposed desktop - HassOS is Virtualbox VM running under Ubuntu Mate (bare metal), Frigate is runnign separately as a Docker container.

[bremby]

HI,

I’m running frigate as a docker container on a dedicated host (not on HA),
where I can find documentation on what parameters/values I can put on docker-compose.yml and what should be on the config.yml file?
what is expected to be a config wide and what is camera specific?

there is a summury/table with available?

regards

[Morphy]

@thundergreen

you can add an iframe to integrate frigate if you run frigate not as addon

This will work locally (on your home network) but to be able to view Frigate externally you would need to set it up for external access (allowing port through the firewall )

[Morphy]

That’s weird you should mention the CPU lowering at night. I’ve noticed the same. I have two corals on a crap Celeron but it chugs along better than the Pi:

Weird how my CPU is still affected by the day/night transition and yours isn’t.

[Morphy]

Have a look on github under documentation and installation.

[bremby]

@Morphy thank you, I did and I’m doing. it can be only an issue with my mindset, but it takes a lot of time and experiments for me to make it work

[bremby]

It sounds that my experiment now works, I’m running Ubuntu 20.10 x64 on a raspberry pi4 with 8gb.
I’m waiting for a coral (when it will be available :(); in mean time I can’t understand how to configure the hardware acceleration.

if I add this

the stream is broken and i see only a green video, the log shows:

ffmpeg sent a broken frame. memoryview assigment: lvalue and rvalue have different structure
[h242_v4l2m2m @ 0xaaaadd85f3c0] could not find a valid device
[h242_v4l2m2m @ 0xaaaadd85f3c0] can’t configure decoder

this is my camera stream

If I comment the entire ffmpeg block, I see the video on frigate webpage and no errors in the log

any suggestions?

regards

[Morphy]

hw acceleration on the rpi4 didnt work for me either so I removed it.

[AussieByrd (Ben)]

To be fair, my sample size is quite small at the moment… Only installed the Coral yesterday, so only about 6 hours of daylight do far, plus overnight (and I’ve been testing some other stuff on my server, contaminating results). Will give an update in a couple of days.

[mspinolo (M)]

Is this supported in the frigate docker image?
I am running USB TPU but that would be a nice upgrade!

Edit: this = mini pcie dual TPU

[Dmitriy_Marchevsky]

Hello!
I’ve installed Frigate, connected my RTSP cam and it’s working great, except for it has never detected object other than person - I have 3 dogs and want to have clips recorded for them as well. I’ve looked through the documentation but did not find any option limiting detection of object tipes. Did I miss anything? Or is it related to the model?