#cameras-archived
1 messages · Page 116 of 1
I get nothing following this https://jellyfin.org/docs/general/administration/hardware-acceleration.html[#hardware-archived](/guild/330944238910963714/channel/672223433907765288/)-acceleration-on-docker-linux and running sudo lshw -c video or vainfo I did find this render:x110:pi I can't add the semicolon after the x without making it an emoji for some reason
@granite copper do you know what the device is for a raspberry pi4b running in docker? I'm running ubuntu server now if that matters.
- /dev/dri/renderD128 # for intel hwaccel, needs to be updated for your hardware
I think it's /dev/video10
Yeah that's the first one listed. I guess that would be it then right? https://dpaste.org/9yOJ1
How much gpu memory do you recommend for a raspberry pi4b 8gb ram? Default is set at 76 right now, should I push it to 128?
I always just do the max. Not sure.
I think it's 256/512 right ? https://github.com/raspberrypi/documentation/blob/develop/documentation/asciidoc/computers/config_txt/memory.adoc it says it's dynamically allocated does that mean it would go from 76 to max if the application required it?
I don't know if I can catch a break trying to migrate to the pi lol. I got hardware acceleration errors, ok turned that off in the configuration and docker compose.. then i got fatal python errors. Seems to be an issue with the tpus? I am using both on the recommended USB powered hub on the 3.0 port. I even tried just using 1 on the USB hub and that still generated the same errors. I even tried using the detectors without the hub by directly connecting and still getting the errors. https://dpaste.org/suY0h
https://dpaste.org/fUNae
It will say edgetpu detected sometimes only to say later on no edge tpu detected. Seems as if this is just not meant to be lol. Plugged them right back into the odroid running home assistant os and they work.. not that I thought the issue was them.
I always just use the official raspbian 64bit build for testing.
does the accuracy of detection vary between coral and cpu? i.e - is coral detection more accuracte than cpu? I have cpu and detection at night is all over the place. cats get detected as birds and vice versa. I got detected as a dog lol. I also understand that the camera only sees 2d so it may not be so accurate at night
So is there a library I should install for the tpu then on Ubuntu? How do I make sure portainer would have access to it.
Accuracy should be the same, but with a TPU it can run detection more frequently. It takes multiple frames to confirm, so it may miss more objects with a CPU.
You can install the libraries from the Coral.ai docs, but all they do is setup permissions on the device. That could be your issue I guess.
The process in the container runs as root, so that hasn't been necessary in my experience
Yea. You can try installing the runtime. It's already bundled in the container but you can follow that tutorial to see if the getting started tutorial works at least.
Sounds good. I'll report back tomorrow hopefully with better news lol.
Do you run the tpu at max? libedgetpu1-max
Yes
Perfect.
I wonder if this is the issue?
The following packages have unmet dependencies:
python3-pycoral : Depends: python3-tflite-runtime (= 2.5.0.post1) but it is not going to be installed
Depends: python3 (< 3.10) but 3.10.4-0ubuntu2 is to be installed
E: Unable to correct problems, you have held broken packages.
pi@ubuntu:~$
so its normal for incorrect detection in the dark? I think shadows also contribute to that. I have attached a screenshot https://imgur.com/a/bjXnpOG
Probably not. The container includes all that stuff itself.
The models are trained on color images, so they are less accurate at night, yes.
Sounds good, I'll try again tomorrow. Have no idea how many times I have tried already lol. All to get a USB hdd and offload the CPU load to the raspberry pi from the odroid and use the hardware acceleration that I can't seem to find out how to use on the n2+
I find them to perform very well on my cameras. It varies a lot based on the camera and the situation.
To clarify, I mean at night too. I very rarely get false positives even at night.
are you using dahua IPC-HDW5442TM-ASE? and did you change any settings in conditions tab?
I am building a unraid server with frigate specifically in mind. would the amd igpu work with frigate? I heard plex has issues
Back again, It still didn't work. I actually noticed frigate not working caused a bus error and made the pi completely unresponsive. If it was a power issue somehow (even though I'm using a powered USB hub) I got a 45w USB c charger connected and still got those errors. I was never able to do the demo because of the errors in installing pycoral. What's odd is that 1 tpu is flashing and the other isn't. This could be my issue. When I ran lsusb yesterday I didn't see Google listed at all, it may be that there's an issue with identification. Even though it identifies it for a few seconds something might be wrong in the way it's identified and used? I'll attempt what was listed there. There's suggestions to passing through the entire port rather than the current way and editing udev rules to identify as a google coral. I tried to see what it identifies as on home assistant os but I can't see it. They are identified immediately though in comparison to running on the pi. https://stackoverflow.com/questions/56632485/coral-google-edge-tpu-usb-accelerator-not-recognized-virtualbox-workaround
I don't know anything about your issue but it's pointless to use a 45w USB c charger for the pi
45w is just achievable through USB-PD at higher voltages for charging purposes
The Pi just uses the 5V, which is probably 3A = 15W (like the Pi 4 stock charger)
Do you know if 12th gen has issues with igpu and frigate?
I don't know. I would look around regarding ffmpeg hardware acceleration and that processor. Frigate just runs ffmpeg to decode.
You should open an issue on github. Maybe someone who actually uses a Pi has seen something similar.
is anyone here using Frigate + DoubleTake?
for Facial Rec?
looking to chat with you if you are and see how im supposed to be using this thing. I've got deepstack doing facial rec via DoubleTake
I needed to really push up shm size. It's relatively stable now but about 120ms or so slower than my odroid n2+ on inferencing. I tried /dev/video10 and the hardware acceleration arguments in the docs for the pi and it crashed immediately again. So I assume video 10 isn't it but how would I find out which one it is? There's a lot.
Probably a driver issue since you aren't using raspbian.
I have seen 11 and 12 referenced too
I have changed the detect resolution to 1080p 5fps to see if cats stop being detected as humans. but I noticed that an event that was detected at 11:25 is still in progress at 12:35. maybe this is a one off but thought of reporting it just if it was a bug. please see screenshot https://imgur.com/a/MIkD06A
Issues should be reported to github, not discord. If you look at the debug view for your camera, you will be able to see whatever frigate is still tracking. You probably have a false positive or are using motion masks wrong.
I want to add latest snapshot of a camera with my alert. Where can i find that. I see it in /media in the web interface but on the filesystem it's not there. Where are they?
I made some changes to try and improve audio support in stream. We still only support mp3 and aac audio, but hopefully we can enable some of the cases that we disabled before. Netatmo, Reolink, and Milesight are some cameras that were affected. Can anyone help test this on the beta?
Hey guys, what happened to the MJPEG integration configuration example? https://www.home-assistant.io/integrations/mjpeg/ is not available anymore to be defined from the configuration.yaml or what? I'm kinda in a rush because the only file I kept for re-installing hass was the configuration.yaml one, and now I'm kinda looking over all the errors and trying to get everything up again /o\
This is the error ERROR (MainThread) [homeassistant.components.camera] The mjpeg platform for the camera integration does not support platform setup. Please remove it from your config.
Adding MJPEG IP Camera to your Home Assistant instance can be done via the user interface
I want to skip that as much possible and continue using the configuration.yaml @hollow ore
I dont want both
Good
I've been using hass for around 5 years, upgrading Python to the latest version possible everything and everything is defined on the configuration.yaml
Well, for the last few years HA has been moving more and more things to the UI, away from YAML
But then this is deprecated now? gone?
Yes
- It was in the release notes under Breaking changes when it happened
- The docs always tell you the truth of how an integration is configured
But you said i could still used yaml only?
Sure, i missuderstood then
I said things - as in in general - are UI or YAML, not both
This integration is UI, so no YAML option exists
i want MJPEG to be YAML only, but then there are no examples for it anymore..
Well, of course not
It's not possible, you can want it all you like, that's not going to change anything
But i see for MQTT https://www.home-assistant.io/integrations/switch.mqtt/#new_format ?
Sure
It's like saying that because your coffee comes in a mug so must your steak
If you really want YAML, it's time to downgrade HA to before that integration moved to the UI
And never upgrade HA again
I wasn't planning to upgrade.. My sdcard crashed and have to install everything and restore the backups (just the configuration.yaml for what i knew)
Oh well
Thanks for the clatifications anyway Sr
HA's config is in far more than that one file
How are backups are to be handled from now on? I'll try to get my home automation working again and at least prevent this to happen in the future
I've put this in cameras since it's the closest I could spot to somewhere to ask about light sensors.
In another building I have various equipments with status LEDs on them. I want to be able to remotely confirm these are OK. I'm thinking of a device with a few self adhesive photodetectors stuck to the front of the LEDs which then will put the relevant lit/unlit signal onto WiFi.
Does anything like this exist? Or would it be a DIY job with esp32 or similar?
@glacial impupdate on that issue from like two weeks ago, just wanted to throw out I'm test driving the wyze cam v3 on the RTSP firmware and it seems to be WAY more stable than the V2's. Like, alot. Haven't experienced a single dropout yet with Blue Iris
FOV and picture quality is marginally better too, but the real bonus will be the starlight sensor. Have to wait until tonight to see the diff
anyone here know if i can get ptz control working for my tuya cams ?
ive been googling for a bit but not much luck
Can you see the live view inside the area card?
what do you mean by that ?
i get output from the camera if that is what youre asking
i just wanted to have ptz control aswell
Why don't you just use an NVR/Software made for it?
any pointers on what to use ?
these are tuya smart ptz cameras
Blue Iris
Just make sure your cameras have an RTSP and/or ONVIF stream
Actually you need ONVIF for PTZ control IIRC
Hey guys, I have an issue with that some streams gets like this in Frigate NVR. Had the issue over multiple versions (1,5 years) and figured I should get it resolved. https://imgur.com/a/4smr4zz Those three cameras are all different types, Axis, Bosch and D-Link. What's the cause? It's so wierd that 10-20% of the top part is perfect then it renders like that below
Anyone running Frigate on a server with dual Xeon CPUs? I'm looking for some advice on setting up hardware acceleration. None of the examples from the website work and I'm not really familiar enough with ffmpeg args to do it myself.
E5-2430s
Unfortunately a GPU may be a little difficult to get in there. I plan to try, but until then I'd like to try the CPU
Wish scalpers weren't asking 8-12x MSRP for Coral TPUs
The coral helps with the inferencing while the "hardware acceleration" is usually referring to hardware decoding
So basically they are helping with different parts. Ideally you'd have both
Good point. It would be nice. I wish I could upgrade these, but unfortunately these are the best CPUs that are compatible with this server
Yeah. You could look into getting a different device. Honestly everyone is focused on the coral since it's plug and play with frigate, but if it's that hard to get people should try one of the other platforms like the Jetson devices or the NCS2. Blake doesn't want to waste time with supporting those as he's working on other features, but it should be easy for someone to just maintain a fork with support for those devices.
I have an Nvidia Jetson Xavier development kit that I'm trying to fix. I may end up running Frigate on that if I ever get it working.
Save my server some effort
Will frigate support h265 in a future release? @granite copper
Frigate already supports h265 for decoding and for all features on browsers that support h265 playback. Frigate isn't the limitation. Browsers don't have h265 support generally and probably never will.
Interesting, so frigate can take an h265 feed and record/use it for detection but for playback step down to h264 on non supported browsers?
No. It can record and use h265 for detection, but playback will fail in browsers that don't support h265.
You have to re-encode h265 to h264 in your ffmpeg args if you must. It's not really practical unless you have a decent gpu.
Probably cheaper to just replace your cameras than buy a GPU.
I see, I was just curious as I'm using blue iris with h265 and once the pr for manual event creation is merged into v.0.12s beta I will migrate completely to frigate so I wanted to know if it had support and to what extent.
My cameras support both
Future versions could support dynamic reencoding of recordings to h264
Interesting, not a big deal I just saw a major bandwidth benefit to using h265 vs h264 with my cameras
Also it would be cool to be able to click on a camera from a birdseye of all cameras to enlarge the camera, if that's possible to do in a future version. I can make a feature request on GitHub.
Actually Chrome 104 (the current beta) supports H265 when started with --enable-features=PlatformHEVCDecoderSupport
Yes H.265 is much better which is why I spent quite some time getting it working in HA. You should use it if you can. Frigate does have a few gaps in H265 support: 1) clips used to work with H.265 before v0.9.0, but once clips merged with recordings they stopped working, so there's probably something missing in the remuxing code and 2) The use of RTMP to restream is not compatible with H.265. The switch from RTMP to RTSP was on the v0.10.0 milestone but must have gotten removed.
For every operating system?
It's something in the Nginx VOD module which I hope to remove in the future. I also plan to dump rtmp and embed rtsp simple server as a replacement.
Yeah, I think it's across operating systems
I tested it on Windows
The support is still behind that flag, and that flag is not even exposed in the chrome://flags yet. I'm hoping it will become standard in a few months.
It's easy to pass that flag when starting the browser from a desktop, but to enable it in android webview is a hassle. I'm hoping the support will remove the need for using exoplayer for H.265 support in the android app
hey guys i wanted to ask for a recommendation for a nvr solution from people with experience. im running HA in Hyper-V, got 2 wifi cameras and ideally would like to do facial recognition but for right now id be happy with motion detection. right now i just set up motioneye but its a bit too light and demands a bit too much effort configuring for things like automations also havent figured out ptz at all with it. also cant trigger high res channel recording with low res channel motion detection. frigate looks awesome but the hardware is impossible to find online and also passthrough to hyper-v seems tricky. zoneminder looks like it can do a lot but also high learning curve. soo yeah im interested in your opinions...
i personally would say start by switching to proxmox and passthrough shouldnt be that bad either
well the reason i went with hyper-v is resource scaling since afaik virtualbox permanently reserves ram. havent looked into proxmox yet, never heard of it actually. will look into it.
proxmox has ballooning ram
so you set the start amount and the max it can expand to
any tips about a nvr solution?
i have heard good thing about iSpy
dude proxmox doesnt run on windows but rather bare metal from what it looks like. im sorry but having tried it and had my problems im really hesitant switching off of windows as a base os
its just a familiarity thing
thats your preference i personally don't want to use anything else for virtualization
plus windows itself is just a power hog
if you want to run multiple things neatly in a container or vm linux and vm's is pretty much the industry standard
dont get this the wrong way im sure its amazing just like i could just run linux bare metal but after spending a month googling how to get the touchpad to work under ubuntu and same for setting up a network share and on and on and on id rather stick with the os where i know how to configure and handle a lot of the stuff i need. im not against learning new stuff but at the end of the day i need it to work and if i dont know what im doing thats unlikely
lets forget hyper-v for a second. frigate for example would run seperatly anyway. so im just looking for a ggod nvr solution that run under windows or can be virtualized under windows
If you are really set on windows, you are probably better off just paying for blue iris
It should feel familiar since everything is UI config. It is very windows like if that's all you are comfortable with.
im sceptical about the features. right now im rethinking how to make frigate work for me
are you running it in docker on windows? what kind of accelerator are you using?
im using it in a docker on unraid, no accelerator and its fine
i have 2 1080p cams for reference
the HA integration is really awesome, the fact that it gives you a sensor for "person detected in this area of this video feed" is incredible for automation
could you elaborate on your video feed used. i heared the coral accelerator is really recommended. i have one 1440p/20fps channel , one 480p/7fps channel and a seperate 1080p camera planned with propbably another 2 channels
theyre both rtsp 20fps at 1080p, in my setup i get out about 5fps I assume because im not using the accelerator
but honestly i dont mind at all, if Corals ever become accessible again ill probably pick one up but for the time being its totally fine
well i have the oppurnity to buy one thats why im rethinking it all 😄
i would definitely do it
hi guys.
I have successfully set up WebRTC and got by 2x Nest and 2x Wyze v3 cams to work with HA. I have set up Picture Glance cards on my dashboard to view all of the feeds, however one thing we miss dearly from both the Wyze and Google Home apps is the ability to zoom in on the live feed. So when we click on the Picture Glance card for one of the feeds, the "viewer" that is opened up doesn't allow us to zoom.
I am wondering if there is something I can use that would allow this? Is there something other than the Picture Glance cards that I should be using to view the feeds.
I have a 3+1 parking space behind my house, 3 next to each other and one across the back of them. If I had a camera facing it from above could I had Home assistant use AI to identify which car is in which spot, not by license plate but by make/model or at least a way to tell who is parked in what slot somehow?
Hello, looking for some help with WebRTC... After update my reolink IP cameras will not work... Has anybody had this same problem? Thanks for any help in advance!
I got a new hikvision camera and I am seeing a lot of these logs in the frigate container log. anyone know why?
[2022-07-07 16:13:30] ffmpeg.FrontYardNorth.record_rtmp ERROR : [flv @ 0x56354d49cdc0] Video codec hevc not compatible with flv
[2022-07-07 16:13:30] ffmpeg.FrontYardNorth.record_rtmp ERROR : Could not write header for output file #1 (incorrect codec parameters ?): Function not implemented
[2022-07-07 16:13:30] ffmpeg.FrontYardNorth.record_rtmp ERROR :
[2022-07-07 16:13:30] watchdog.FrontYardNorth INFO : Terminating the existing ffmpeg process...
[2022-07-07 16:13:30] watchdog.FrontYardNorth INFO : Waiting for ffmpeg to exit gracefully...
[2022-07-07 16:13:35] frigate.record WARNING : Discarding a corrupt recording segment: FrontYardNorth-20220707161332.mp4
this is my config for the specific camera
FrontYardNorth:
ffmpeg:
inputs:
- path: rtsp://admin:1234567a@192.168.11.22:554/Streaming/Channels/101/
roles:
- record
- rtmp
- path: rtsp://admin:1234567a@192.168.11.22:554/Streaming/Channels/103/
roles:
- detect
detect:
height: 720
width: 1280
fps: 10```
NM, I forgot to set h.264 in the camera. its all good now
Yeah you could do that if you're familiar with training a classifier, need to capture a training set of images though
Otherwise you could try using some pre trained model or general vision API, but those are more general/complex solutions and not tailored towards your camera orientation, so it might be overkill and still have worse results. Plus you'd still have to do the same work of cutting up the image into the different spots either way, and with the generalized solution you'll need to translate the generic results (e.g. blue car, red SUV) into your own results (wife's car, unknown vehicle)
Basically it's doable, but you'll probably need to do some work
Hi everybody whixh video dorrbell with intercom and bell would you suggest? yum
I'm using an Amcrest AD-110 but there's a newer AD-140 available. Others may be using different ones but I can't speak on them.
@granite copper - What does the v0.11 feature "Send to Frigate+" do? I have a Frigate+ account created, and API key set for my frigate container. Just wondering what sending a clip to frigate+ will do
It uploads the snapshot from the event to your Frigate+ account so you can add labels there. It does the exact same thing as uploading an image directly on Frigate+ manually.
Gotcha. Looks like it's hanging on 0.11.0-b36b635 . I get a log message saying that PLUS_API_KEY is not set, but I exec into the container and I can see it in the output of env
maybe it's something that just isn't ready yet?
How did you set it?
in docker-compose.yml
That should work. There are lots uploads every day.
could be my syntax. When I use the format PLUS_API_KEY="blahblahblah:blahblahblah" should I have quotes around the key? or should it just be in there without quotes?
I wasn't sure if the ":" would make the parser mad, so I put quotes around it
I don't know off the top of my head what version that is. Is that a published beta tag?
If it doesn't match the format, I think there is a log message on startup.
So I discovered something neat with the Wyze cams and the docker wyze bridge. For some reason, the wyze docker bridge stabilizes the feed almost completely, even if you're using the RTSP feed direction from the camera and not through the bridge. Now, docker wyze bridge does need the internet to authenticate at first boot, but you can cut the internet off after authentication and it'll still work after.
What I'd like to know is what exactly does the wyze docker bridge do that stabilizes the feed so well.
I grabbed the tag from docker hub. It was the most recent build as of 2 days ago.
Does it work with the latest beta tag?
is that beta5?
@granite copper ok, so upload to frigate+ does not work on blakeblackshear/frigate:0.11.0-b36b635 , Digest:sha256:d6c20979d59f49557c9828cb2e6a8088e034cd0477e47591829f0c913e3b405e (amd64) , but does work on beta5 (blakeblackshear/frigate:0.11.0-beta5, Digest:sha256:2208dd73a9eba53ffdc53c52d1e2fca3f3619b0a4804a879c3c169e4c050779a (amd64).
the b36b635 build gives the error about not being able to find the PLUS_API_KEY
Is there a program or integration that you would recommend like deepstacks or frigate that can do this?
Ok. I will have to see what changed to cause that.
Thank you!
Also, when is frigate+ going to be available? I'd like to sign up to support the project, if nothing else
I appreciate that. I think I am getting close to having enough user submitted examples to train better models. I already have had success training models directly on the pixel formats that ffmpeg outputs which will avoid color space conversion overhead. If I do have enough examples, I am hoping to have the first models available in a couple months. Once those are available, I will enable the subscription links. In the meantime, the best way to support the project is via github sponsors.
Awesome, thanks!
By the way, what is the best way to submit false positives? I have constant problems with clouds being reflected off of car windshields in the driveway as people
Hello, I am trying to setup frigate but I am lacking docker experience. I am trying to setup a dedicated frigate container via docker-compose. my config file is on c:\docker\frigate\config.yml. What do I need to enter to get this reflected in volumes:
volumes:
- /etc/localtime:/etc/localtime:ro
- /path/to/your/config.yml:/config/config.yml:ro
I am unable to find the correct syntax for this
this seems to expect linux paths, what's the correct syntax for a Windows based file?
You can leave out the link to /etc/localtime. For your config.yml, google says this: https://stackoverflow.com/questions/41334021/volume-binding-using-docker-compose-on-windows
thanks, I will remove it. unfortunately that thread is focused towards the old WSL which leverages hyper-v, today WSL2 doesn't use virtualization and is not offering "file sharing" options but thanks for the pointer anyway, I do really appreciate all the help I can get
can't you just run docker on ubuntu with WSL2? Then you can use the native linux syntax and not have to worry about being a corner case
I thought that's what I am trying with docker-compose?
oh you mean run docker from inside ubuntu and not from windows?
yeah, docker on ubuntu inside of WSL2 instead of docker on windows
I think it'll make your life easier in the long-run
I have an USB coral connected to Windows and I am not sure I can nest it down
I know I can connect it into docker on WSL, not sure how to get it into docker in Ubuntu though
ok, got that part figured out, thank you, now on to the next challenge...
You would upload the image and verify the image with zero labels, assuming none of the objects are present. I plan to add a more specific false positive report feature later.
I uploaded a couple of images and labeled the cars, but didn't label any "person" objects where they were currently being flagged. Is that the correct process?
Yes
ok, finally got frigate running and detecting the coral. I am getting a " | F driver/usb/usb_driver.cc:1148] HandleQueuedBulkIn transfer in failed. Invalid argument: USB transfer error 4 [LibUsbDataInCallback]" error. have not found infos on USB transfer error 4, only 2 and 5. does anyone know what 4 means?
In my experience, this is typically related to power issues on the USB ports. The Coral is very power hungry and some devices don't offer enough amps on the ports. The docs recommend a powered USB hub if you are on a Pi. It could also be some other issue if you are passing it through to a VM.
ok, thank you. not passing through a VM but WSL2. do you have any guidance on debugging if this related to this or lack of power?
Are there any other USB devices plugged in? You could try unplugging those.
I really haven't used windows in forever, so I am not sure. You could search github issues for other users who might be using it.
will do, thanks
Is WSL2 actually just a VM?
no, WSL2 has no Hyper-V VMs involved
it's native integration
(it does offer a legacy Hyper-V option as well but that's basically outdated)
Has anyone in australia managed to get coral from rs components? I have it on backorder and They keep pushing the date further
As I mentioned you'll probably have to do most of this on your own. The use of any external tools is really secondary to how you design the rest of the process. If you are already familiar with this kind of stuff it might take you a few days. If not, it may take a month or more as you'll have to learn as you go.
update: i had the config set wrong at it was limiting it to 5fps. theyre both running fullspeed now
@rain geode WSL2 is a VM, WSL1 was the native integration approach. https://docs.microsoft.com/en-us/windows/wsl/compare-versions
did any of you guys saw the webrtc integration broken after the last two updates? I had 3 cameras and both doesn't work anymore
I'm using latest, I think, and is working for me.
So, your card is similar to this?
type: custom:webrtc-camera
url: rtsp://192.168.1.5:8554
My typical usage is to leverage webrtc from the frigate card, where I point WebRTC at Cameras registered via the Amcrest integration.
I just confirmed that accessing my cameras using the form you posted works correctly.
Hey folks! Is there any progress towards bringing Homekit Secure Video (HSV) to HA or is the idea to always use scrypted for that?
Is there a way to delete recording for only one camera via frigate gui?
Hey, any idea why frigate in HA is driving crazy in console? I don't get a video stream of my 4 cameras: https://paste.pics/HK45Y I first have to reload my chrome browser window to get the stream working again.
I am using the frigate-card
I always get an delayed stream when I access my cameras via frigate. Is there a recommended way to access an up to date live feed of each camera?
It’s maybe like 30 seconds behind real-time.
im wondering, does it help to have a dedicated videocard on my HA system when using webrtc?
I have 8 coral pci-e versions available. Selling at cost plus shipping. Dm if interested. Thanks
@granite copper I got a warning in the logs to stop using qsv because qsv is deprecated and when I switched to it I got Unrecognized hwaccel: -hwaccel_output_format qsv
ERROR : WARNING: defaulting hwaccel_output_format to qsv for compatibility with old commandlines. This behaviour is DEPRECATED and will be removed in the future. Please explicitly set "-hwaccel_output_format qsv"
Is there a way to show Amazon Echo Show Camera on HA Frontend?
Any way to change the resolution of the ring camera when doing a camera.snapshot? I'm getting a resolution of 80 x 80 pixels which isn't very useful.
What's frigate performance like using quicksync on a i5-8500t?
And can it record to a network share?
I bought a few OptiPlex 5060Ms in a surplus auction and am thinking about moving my Frigate instance to it
Can frigate detection be set at different resolutions per camera?
Have one 1080P camera and one 2560x1920.
Recording notifications fail on the larger resolution one
yes
ill give it another shot, threw a ton of errors at me last time i tried.
frigate wouldnt start after i tried that
are you getting ffmpeg errors?
i was, hang on, let me try again now
shocker, starting fine now :-p
i think i may have put a trailing space on detect
I am trying to figure out how big hdd do I need. I know its hard to say but would 6TB ball park would be enough to hold footage from 8 x 4k cameras for 7 days? I am using frigate as my NVR and have to get storage before I get any more cameras
boy too many options in the frigate card, i'm lost , i used to have an event carousel on the bottom of camera feed , anyone know how to get it back ? its now hidden in an leftside drawer which is annoying ...
@prime notch try https://github.com/AlexxIT/WebRTC/issues/318#issuecomment-1178980046
Thanks a lot 😀 it seems that they find the fix, I had to restore on previous update to keep it working otherways
is this the correct way to create mask for "persons"?
track:
- person
- car
- car
- bus
- dog
- cat
filters:
person:
mask: 971,477,1024,490,1049,559,1005,600,904,501
changed to this
person:
mask:
- 971,477,1024,490,1049,559,1005,600,904,501```
but when I go to debug and turn on masks, I dont see the mask
Cross thread from #integrations-archived but quick question - Has anyone got the new Google Nest Camera to show up in HA? As they need to be configured within Google Home they dont show up under my nest integration
Any way to integrate dahua dvr?
And another thing, this is my network setup [https://imgur.com/a/XS9XXnm] Is there any way that i can acces the dvr? (also i need to acces the devices that are connected to both routers)
also i can acces the dvr via my personal pc
Do you have Router2 connected by the WAN or LAN ports to Router1?
HA can reach the DVR, though whether anything on the Router1 network can reach HA is another question
lan
Then it can access the DVR, since they're all on the same network
found something but i don t understand what i should complete on "host" field
The IP of the DVR?
Anyway i found how to make it, ty
can someone confirm if I can use hardware acceleration with frigate and ryzen 5700g?
I have it working on my 5600g
how many cameras? what is your cpu usage?
running ubuntu 20.04, but I had to update the kernel using Hardware Enablement Stack
I am on unraid so probably different process would apply
6 cameras
do people pretty much stay with the 3 frigate suggeted cameras?
anything that ffmpeg can consume works... rtsp camera's might be the easiest... I don't think people are restricted much at all.
Any recommendations for a cctv kit? Hikvision, reolink?
I usually just look at the youtube videos by people, frigate itself has some suggestions.
i was trying to get an idea if there are any decent ones for sale today or tommorrow
my mask does not work for a person. frigate thinks that the tap is a person. can someone check what I did wrong please https://pastebin.com/MHKkmwq0
"Keep in mind that this mask only prevents motion from being detected and does not prevent objects from being detected if object detection was started due to motion in unmasked areas." https://docs.frigate.video/configuration/masks
so the below code will still detect person in the masked area if the motion was in a different area
person:
mask:
- 971,477,1024,490,1049,559,1005,600,904,501```
That object mask shouldn't, but note that it is only based on the bottom center of the bounding box. It doesn't matter if 99% of it overlaps. Only the bottom center of the bounding box matters.
what's a good cheapish camera for HA right now, ~50 euro, mains powered, live feed and onboard sensor (zigbee, zwave) for motion detection?
I used to run android ip webcams from some old phones previously, maybe there's something better?
got a cheap TAPO C210 during Prime Day. EUR 39. You need an app to set it up (is there anything which does not need an app?) but afterwards it works fine with HA.
thanks, looks pretty good, i can even get 2 for ~50
Hello
I am looking for an easily integrated interior camera, which records on my SSD.
I saw this Reolink 5MP PTZ Camera
It looks good
Would that be suitable in your opinion.
Thank you for your help
Mike
Reolink is ok, but every dev in HA/frigate has experienced issues with them since they use a buggy/old live555 stack.
Also, the folks at ipcamtalk will tell you how they tend to overhype themselves.
But they are decent cameras once you get them to work, and they have a good feature set.
Hikvision/dahua and their rebrands are the better choices, but the good models can be more expensive.
Btw, reolink is working on updating the live555 libraries in their firmware. They're currently using a stack from 2013 or so and there are multiple CVEs associated with it.
However, the new firmwares are still alpha or beta and are quite buggy.
Anyway, I'm sure reolink has great sales and marketing as I always see people posting here about them.
My ONVIF camera has a several motion events and signals and what not... how do I set it up so that it records on HA if there is motion?
@rustic rampart
thank you
maybe he. there would be a model that you suggest
I bought the EC60 indoor cameras from Kasa before knowing HA was a thing… best way to get the data from the sensors onto HA? As far as I know it is not RTSP supported. Mainly just want to use to push notifications to my phone when motion is detected.
I bought this ezviz camera and it works fine https://www.ezviz.com/us/product/C6CN-Pro/24377 . Ezviz's product lineup is a bit of a mess though, with many different but similar sounding numbers, and I can't speak for the other models. Also, availability seems limited to certain models especially in the US.
Otherwise, you can try the amcrest 1051
Looks decent although I don't have direct experience with it
You can make an automation in HA
Use the motion sensor as the Trigger
And then Call Service -> Camera: Record in the actions
I was able to get rotation to work, but it doesn't work in the iOS app
Awesome! I don't care about the iOS app. 😁
Well we try to maintain a consistent experience but there's only so much we can do. I may consider adding it in 2022.8 or 2022.9
Hi all. do I need to 'install' ffmpeg to get it working on a HA OS VM?
No
I'm trying to add rtsp streams to my HA by adding the camera details to configuration.yaml
That's not supported as of ... 2022.6?
I added stream: and camera: (plus all the rtsp details). the picture card can show the still image but when clicked... doesn't really show the stream
oh. didn't know that. what's the way to add streams
Following the instructions in the docs?
The camera is added in the UI
stream is included by default if you have default_config (see https://www.home-assistant.io/integrations/stream), so nothing else is required
ok. I'm reverting the changes in the yaml and trying to do it via ui
still having the same issue. I provided the stream source. the still image gets generated but when I click on it to play, the player just keeps flickering
Is there anything in the log file?
If you're having problems with your updates to your configuration:
- Check the troubleshooting steps
- Check your log file - remembering you may need to set logger to
infoordebug - Explain what the problem you're having is - sharing configuration, errors, and logs
nothing in log file but i just saw the text flickering on the player. "error with media stream contents"
could it be the codec of the stream?
What kind of camera do you have?
For some camera types, there was a regression introduced in 2022.7.0 which should be fixed as of 2022.7.5 (just released an hour or two ago)
I think I might have found the issue. found a post that says the browsers do not support H265 streams.
I've just changed the encoding on the camera
Yeah, or you can use a different browser
Safari supports it
Edge has supported it if you install an extension
Chrome beta supports it if you enable it
ok. now it's fixed... sigh... kept thinking it was on the HA side
I'm using a low quality sub channel for viewing so I don't mind changing it to H264 for compatibility
thanks for the help.
Cool, glad you got it to work
anybody here use generic cameras in HA and view their rtsp cameras in HA via a browser in Linux ?
Don't ask to ask, just ask your question. Then people can answer when they're around.
When you do ask a question, try to provide as much background detail as possible. Ask yourself these questions first so that others don't have to:
- What version of the Home Assistant are you running? (remember, last isn't a version)
- What exactly are you trying to do that won't work?
- Is the problem uniform or erratic?
- What's the exact error message?
- When did it arise?
- What exactly don't you "get"?
- Can you share sample code, ideally with line errors where the error occurs?
I have my cctv camers in HA with rtsp stram using genreric intergration in HA. I windows i had to install HVEC from windows store to get them to strem inside HA in edge. I think you can install a browser extention but never needed to but they worked correctly. I have moved over to Linux and using same browser my cameras are flickering when trying to stream. As normal they play in VLC but not in HA in Linux i cannot find or do not know as anoob in LInux how to get it working. I wondered if anybody here had solved this issue and could help me please
Read the conversation above this
Home Assistant 2022.7.5
Supervisor 2022.07.0
Operating System 8.2
Frontend 20220707.1 - latest
This is a short video of the issue
It is normal across all platforms i beleive due to H265 licenses i think but with correct codecs installed you can get it working. But as a noob in Linux i have tried different things from Google searches but still can not get it working. I thought i would as here as it may have impacted other users
Problem is constant , no error message
No errors logs generated from what i can see
I have tried download diagnostics in the intergration but it does not work for me !
Which browser are you using?
Try using Chrome 104. It's the current beta: https://www.google.com/chrome/beta/?platform=linux
Launch it along with the option "--enable-features=PlatformHEVCDecoderSupport"
a wow okay found it thanks! I looked at the automation thing but didn't figure out that there is a camera service to call.
If you use that feature, you'll probably want to enable preload stream as well. That way you can use lookback when you get a motion event
That will let you record starting before the actual motion happened
HI i was on Edge Dev
Edge Dev is 105...you should be able to do it there too
Open it with that option I mentioned
Try opening the browser from the command line
With that option I mentioned afterwards
sorry to be noob what is the command to do that ?
Which linux distro / desktop are you using?
Can you create a shortcut on the desktop?
Can you open the shortcut and see if there is something that says command?
It's not exposed in there yet. You have to start up the browser with this option
so what do i type
I haven't used Zorin before. But I'm guessing it should be straightforward. Do you start Edge with an icon?
If so, right click on the icon and see if you can edit the properties
nope nothing like that
its based on Ubuntu
I get show details but it opens the software store
What does Open With say?
Can you edit that file as a text file?
Just text editor
Yeah that's fine
Just edit the last line that says Exec
Add "--enable-features=PlatformHEVCDecoderSupport" to the end of it (without quotes)
like this https://i.imgur.com/P8pg92d.png
must have done it wrong they are still not loading
Exec=--enable-features=PlatformHEVCDecoderSupport
is that correct
My ONVIF "lechange" camera has a "cell motion detection", "motion alarm", "global scene change", and "digital input"... and when I walk in front of the camera only the "digital input" comes on. What is the difference between all these?
wait and how do I make it stop? I'm not sure how I can see if it's still running. I added a condition on the trigger, hoping it'd stop running when the condition ended, but maybe I need a seperate stop action?
Basically, the input is off, but I don't see a new media file yet
There should be a duration in the service call
I think the default is 30 seconds
Try just calling the service directly first to test
like this Exec=/usr/bin/microsoft-edge-dev --inprivate --enable-features=PlatformHEVCDecoderSupport
@next walrus yes
@foggy imp Go to Developer tools -> services and try to run a recording
Using camera: record
Yea, what I would like is that it records until the motion detection stops giving a signal, or will it just retrigger if it's still recording
ok lemme try that
Hmm, that exact functionality isn't currently possible as the duration is fixed when the service is called.
But you can get close enough..just have it record another clip if there's subsequent motion
still not working
hmm I get a permission error on the path
Are you starting Edge with that desktop link?
on desktop link yes right click open
Yeah you need to make sure you have a path you can write to such as the /media directory
ah I was going to ask.. what is a sensible place to write to on this thing
What CPU do you have?
GenuineIntel
CPU family: 6
Model: 158
Model name: Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz
Stepping: 9
CPU MHz: 2609.057
heyyy works, so.. ok only fixed length recording...
See if you can install the following packages: "i965-va-driver-shaders libva-drm2 libva-x11-2"
done
hmmmm how do add the date to the filename or something like that?
still not working
Yeah, see if you can do what you need to with automations. You could just make them like 1-5 minute clips. It's unwieldy to deal with really long clips anyway
https://www.home-assistant.io/integrations/camera#service-record supports #templates-archived in the filename
IMO you're likely a lot better off looking at an NVR
What happens when you go to chrome://gpu in the browser?
Hmm, try two more things
ok
- Try installing libde265
how
- Try adding "--enable-features=VaapiVideoDecoder" to the edge.desktop file you edited before
Exec=/usr/bin/microsoft-edge-dev --inprivate --enable-features=PlatformHEVCDecoderSupport --enable-features=VaapiVideoDecoder
Yeah..does that change what chrome://gpu displays?
how do i install libde265
Actually, just try the package "ubuntu-restricted-extras"
Oh you also have a GTX1050 in there. You can also install mesa-va-drivers
But try to get hardware decoding enabled. Maybe go to chrome://flags/#enable-accelerated-video-decode
I have to go now, sorry I couldn't help more. I'll follow up with you later. Maybe DM is better.
ok cheers when is best next week ?
lol, there is a huge delay between the motion signal and the recording. I went to wave to the camera to test, and you barely see me slam the door
also... it's somehow only showing the substream, which is the low resolution one
hey guys! Does anyone use Frigate? Any tips on how to make the preview feed a better resolution? As it's a bit of a shame of having it set as the 'detect' feed (which I've made very low)
Use lookback
How do i get motion status to work on a picture-entity card of video camera. Camera is working but always says idle.
why do the previews of cameras time out after a couple minutes or so?
like I have to reload the UI for the view to be live again even though it is already selected as live
they all freeze on one frame
@next walrus Sorry, I tried and I couldn't get it to work either. I think it's still a WIP for chromium on linux (although you can find builds with built in support). See this issue: https://bugs.chromium.org/p/chromium/issues/detail?id=1326407
Second time I tried it the delay was almost zero, so it's probably fine. It says something about having to have a running stream... I don't really want it to be streaming 24/7 if there is no motion. Uh... is it always streaming? Does it keep the stream open if you use this option? Or do you need to configure the stream to be always on for this to work? (you da real MVP of this channel it seems, thanks so much)
If you don't have preload stream on, you can't use lookback, so you can only record after you get the motion event. The delay here is not due to the hls stream itself, but probably due to the latency from the motion event. What kind of camera do you have?
Hello. I am looking for an outdoor camera that I could use with HA, does not depend on any subscription to chinese cloud, and would allow for local storage. Would have a recommendation ?
I used to have a cam that exposed a mjpg stream that was easy to make a storage of with a basic script and wget/curl (but that cam is now dead)
Any RTSP/ONVIF capable camera will do
Dahua and Hikvision are regularly recommended brands
I have one that has rtsp but work horribly. I'll look into these brands then
No Dahua on (French) Amazon, very limited choice for hikvision. Lots of reolink however. Do these have a good reputation (quality and HA-wise) ?
thanks
Hello all, I dont want to spam this channel with all this text so i think its appropriate to reference the link. Its around Security cameras + BlueIris + HA and some overlap and making sure what im gonna put time into configurations makes sense and will work.
https://www.reddit.com/r/homeassistant/comments/w1alo0/clearing_up_security_camera_blueiris_ha/
hi, does anyone know why mjpeg cams freeze after like a minute or so in the UI?
feed resumes after I switch to a different view in HA and back to the cams
Imou new bullet. The on-camera stream seems to start earlier, but maybe that's just because it uses loopback locally.
Overall quite happy with this one. @granite torrent the website is all cloud this app that, but you can actually disable it in the web interface and use it either with an SD card, NAS, or ONVIF.
Yeah it probably has a circular buffer on board. That's essentially what preload stream is doing too, but that means you have to be xferring over the network 24/7 and it seems like you want to avoid that.
Looking at using a 6100t, 4gb ram. Anyone have experience with this running frigate as NVR?
Hey I was thinking of installing frigate on a PC and hooking up some cameras and found used 2xUniFi G3 1080p cameras for 60 EUR em... do I get them? Especially if I do not plan on using ubiquity protect? My concern is low light it looks like they have som IR features will that still work?
Depends on how many cameras, I'd guess. Not sure 6100t can do many. I run frigate with an edge TPU doing the object detection and a video card doing the transcoding... frigate container uses 5+GB for my 6 cameras.
3 cameras and hopeful to get the usb tpu accelerator soon that's on back order.
5GB of ram?
are you doing detection on higher resolution channels?
I'm running it on 5 cameras - 1080 on all of them, and I believe I'm somewhere between 2 and 3GB...
Hey Guys, I cant get frigate to stop detecting a tap as a person and bombarding me with notifications. what am I doing wrong with my object filter?
ffmpeg:
inputs:
- path: rtsp://admin:1234567a@192.168.11.3:554/video.h264/Streaming/Channels/101/
roles:
- record
- path: rtsp://admin:1234567a@192.168.11.3:554/video.h264/Streaming/Channels/103/
roles:
- detect
- rtmp
objects:
filters:
person:
mask:
- 971,477,1024,490,1049,559,1005,600,904,501
Hey guys I have an improx dome ptz camera that I can only control the ptz with Tinycam pro. I'd like help getting it working on home assistant if possible, I planned to have it at least use the presets
My blink doorbell camera just has a broken image
Most of my detection is less than 1080 actually. I record every second of all cams at high res, however.
hmm.... yes mine is the same. 24/7 recording.
Have you made any changes (increases) to the default segment lengths or event max seconds? If so that could increase the amount of RAM used as it would need to store more video in memory.
Filter masks are based on the bottom center of the bounding box. It doesn't matter if any other part of the bounding box overlaps with the mask.
I don't think so. Other variables I could think of: I've got 3 systems with a browser always open to a HA dashboard that shows snapshots from all the cameras (updated every 10s), so HA is getting updated snaps from Frigate somewhat frequently. Frigate's python process is using 1.6GB, and the coral detector processs is using 1.3GB, nginx 180MB, and each ffmpeg is about 160MB.
I give it lots of tmpfs cache, but it's only using like 44MB of that.
hey guys, I have a few cameras and was using shinobi, but the streams were too unreliable
Now I've tried installing frigate (don't have a coral and not planning on getting one) but i'd really like to have it as standalone so that HA runs independently from frigate.
I've tried 4 or 5 online how tos but i can't for the life of me get it to work
Is there any easy way to install it on a bare ubuntu server (inside proxmox) ?
I know, but that is not easy enough for me it seems, i don't know what to do with half of the info there..
it runs on the hassos vm but havent used it myself
to me it just seems like half the info I need to install it is missing from the docs ?
- Install Docker and Compose
- Create the compose file as shown, update to suit
docker compose pull && docker compose up -d
Am I right in assuming the up part of the compose command points to the file just created ?
docker-compose.yml would be correct ?
Yes, and yes
hmm.. getting a version error, althout I copy pasted that part from the docs
ERROR: Version in "./docker-compose.yml" is unsupported. You might be seeing this error because you're using the wrong Compose file version.
Check that you have a current version of Docker and Compose
I just installed it ? shouldnt that then always be the latest version automatically ?
Well, depends on how you installed it
If you have an old outdated OS and installed from the OS packages...
i've done apt-get update and upgrade but nothing new was found
seems to work with verison 3.3 but that gives me a tar error ?
What's your OS?
debian
Debian ... which release?
10 turnkey core 16.1.1
how can I stop the tap from being detected as person? here is the snapshot https://imgur.com/a/I53LXye
For static objects that you can't use min/max area filters for, you should use a small object filter mask for the bottom center of the object. https://imgur.com/a/ujmhCLB
are you saying that I should put a person filter for the area that you have drawn?
Yes
Guys, i keep trying use one of my new IP cams the same way i used to have the other ones working but i keep getting this message
using the custom:webrtc-camera card
It seems that i can watch the live feed through VLC, but HA doesn't show anything except a black screen
so is there anything i can do to deal with the codec?
Hello, can I use a frigate camera to send a notification to the mobile app with a camera thumbnail without triggering it by a frigate event? I want to show the picture on a doorbell ring that is unconnected to frigate
I tried the dynamic notification example from here: https://companion.home-assistant.io/docs/notifications/dynamic-content but it's not generating a thumbnail
trying to create an automation with a unifi protect camera that with it detects an object that is a person to turn on a light. I can't figure it out. Anyone have such a thing setup?
I'm trying to install ffmpeg5 on my ubuntu hirsute server to make frigate work but I'm getting the error
The repository 'http://ppa.launchpad.net/savoury1/ffmpeg5/ubuntu hirsute Release' does not have a Release file.
Is there a way I can still install it since it seems to be a requirement for frigate (trying to get my camera's to work ..)
It's in the container. No need to install anything on the host.
Has anyone had a reolink camera overheat? I've got an rlc-842a, and the sucker shuts down around 1:30-2pm PT every day this summer when the sun hits it (the cameras face west). It pops back up later that night when it cools down.
swapping out cables, ports on the switch, and resetting it doesnt help.
sounds like you've narrowed it down.. can you see any logs on the camera itself?
https://www.reddit.com/r/reolinkcam/comments/w2w637/reolink_rlc842a_overheating/ this pops up on reddit but seems that is you lol
i'd rma it at this point and not waste time doing any other troubleshooting
Yeah, thats me
@vivid wharf they wnat me to do some testing to troubleshoot the problem, but yeah, RMA is kinda the next step IMHO
kinda crap for them to want that
expecting more than swapping cables, poe port etc isnt really cool
it'll get deinstalled once to go in a box lol
i used to work for a shitty security camera company deploying chinese geovision cameras lol
Hello, can I use a frigate camera to send a notification to the mobile app with a camera thumbnail without triggering it by a frigate event? I want to show the picture on a doorbell ring that is unconnected to frigate
I tried the dynamic notification example from here: https://companion.home-assistant.io/docs/notifications/dynamic-content but it's not generating a thumbnail
soo, after a lot of "learning" I managed to setup frigate ..
now though I'm having trouble getting rtmp to work
I keep getting errors if I enable it
[2022-07-22 00:15:38] ffmpeg.garage.detect ERROR : [flv @ 0x55cdeded2ac0] FLV does not support sample rate 16000, choose from (44100, 22050, 11025)
[2022-07-22 00:15:38] ffmpeg.garage.detect ERROR : [flv @ 0x55cdeded2ac0] Audio codec mp3 not compatible with flv
[2022-07-22 00:15:38] ffmpeg.garage.detect ERROR : Could not write header for output file #0 (incorrect codec parameters ?): Function not implemented
[2022-07-22 00:15:38] ffmpeg.garage.detect ERROR : ```
I have no ffmpeg params in my config.yml
Can you change your camera to use AAC for audio?
Oh wow.. sometimes I amaze even myself ...
It was simply changing the bitrate of the audio output that fixed it
I'm sorry for wasting your time
I kept looking at the video bitrate and didnt make the link to the audio error
The large majority of security cams including the better ones are from Chinese companies, including hikvision and dahua. Reolink, foscam, and ezviz are from China too, and Amcrest distributes mostly dahua products. Axis and mobotix are notable exceptions.
I know that
Ic, just wondering why you qualified geovision with Chinese
I assumed you were trying to say they were crappy by association
There was actually a bug if the stream was already active (if you viewed it recently or if preload stream was on) and used a very small or 0 lookback, but that should be resolved in https://github.com/home-assistant/core/pull/75580 .