Frigate or BlueIris?
Discussion
I've got some major overhaul on the horizon for my main home server. It's a TrueNAS Scale box.
For the best part of a decade I've been running BlueIris for my CCTV. First on a dedicated machine, later as a Windows VM on the TNS box. I've been aware of Frigate for some time, and as these upgrades mean I am going to have to nix the BI VM and start again I am considering running Frigate in a Docker container and ditching BI.
Has anyone else made the change from one to the other? Any observations?
For the best part of a decade I've been running BlueIris for my CCTV. First on a dedicated machine, later as a Windows VM on the TNS box. I've been aware of Frigate for some time, and as these upgrades mean I am going to have to nix the BI VM and start again I am considering running Frigate in a Docker container and ditching BI.
Has anyone else made the change from one to the other? Any observations?
Not made the change, but I have been running frigate in a docker container for a few years now, its been faultless. I run 10 cameras of which 7 have detection. You will need some kind of TPU for it to work well. I find facial detection does not work well without high res streams to work with. But everything else is great. If you run home assistant you can do a lot of stuff with it. I use it to turn on my outside lights, don't get any false triggers now, like I used to with PIR sensors.
Interesting, thanks.
Do you run Frigate in a container inside HA or are they separate entities?
I ask because I was considering reviving my old HA set up too, which I had been running in an Ubuntu server VM. But might run it as a docker container in TNS also instead.
Presumably the TPU is only required for AI stuff rather than "traditional" triggering? Which one do you use?
Do you run Frigate in a container inside HA or are they separate entities?
I ask because I was considering reviving my old HA set up too, which I had been running in an Ubuntu server VM. But might run it as a docker container in TNS also instead.
Presumably the TPU is only required for AI stuff rather than "traditional" triggering? Which one do you use?
I ran BI on a dedicated machine for a few years, then moved it to frigate in a docker container on my NAS a couple of years back. When I started off it was a bit feature limited, but they are adding stuff very quickly.
The learning curve is a bit steep as you have to manually define cameras, rather than BI which will query the camera and auto-add it, but once setup it's pretty straightforwards. The integration into Home Assistant is really nice - but other than displaying camera feeds and alerts I've not really dug into it that much.
I'm running 10 cameras 24/7 recording (just over 3 weeks storage on a dedicated HDD) on a pretty old AMD Ryzen 3 2200 which also hosts a whole load of other stuff. I added a cheap USB Coral TPU which does the AI inference - I don't think they are the recommended option any more as there's more powerful TPUs, or just using a GFX card. You can just have it record 24/7 with no AI, but the recommendation is to use the AI side of things to classify events rather than just using motion detection - so it will trigger if it detects a person / animal / car / whatever rather than just the trees wafting or heavy rain which really reduces FPs. You can define known people, cars etc - I set the cars up a while back and it does a "reasonable" job of identifying which car is which, but it probably needs more training than I have given it. You can also do things like plate recognition if you wanted.
I've not had any stability issues with it, it's been pretty reliable.
The learning curve is a bit steep as you have to manually define cameras, rather than BI which will query the camera and auto-add it, but once setup it's pretty straightforwards. The integration into Home Assistant is really nice - but other than displaying camera feeds and alerts I've not really dug into it that much.
I'm running 10 cameras 24/7 recording (just over 3 weeks storage on a dedicated HDD) on a pretty old AMD Ryzen 3 2200 which also hosts a whole load of other stuff. I added a cheap USB Coral TPU which does the AI inference - I don't think they are the recommended option any more as there's more powerful TPUs, or just using a GFX card. You can just have it record 24/7 with no AI, but the recommendation is to use the AI side of things to classify events rather than just using motion detection - so it will trigger if it detects a person / animal / car / whatever rather than just the trees wafting or heavy rain which really reduces FPs. You can define known people, cars etc - I set the cars up a while back and it does a "reasonable" job of identifying which car is which, but it probably needs more training than I have given it. You can also do things like plate recognition if you wanted.
I've not had any stability issues with it, it's been pretty reliable.
Edited by random_username on Sunday 15th November 17:43
I might have a technical barrier; I thought I'd be alright without adding an NPU, but from all the documentation it's not recommended even if you don't do AI analysis. I'll have to find a Hailo 8 PCIE card (or a suitable M2 adaptor for a PCIE slot).
Running a Ryzen 5, Frigate doesn't use the igpu from AMD by the looks of things.
More investigation required!
Thanks for all the experiences and advice!
Running a Ryzen 5, Frigate doesn't use the igpu from AMD by the looks of things.
More investigation required!
Thanks for all the experiences and advice!
I run Frigate (inside Proxmox), works well. I think Intel is a bit easier, but it should run on AMD as well. I've stopped using my Coral because the Intel support is now so good.
If you're wedded to Truenas Scale, then you don't need Proxmox, which removes one layer of fiddly virtualisation.
If you're wedded to Truenas Scale, then you don't need Proxmox, which removes one layer of fiddly virtualisation.
After some preliminary testing where my 10 gig NIC didn't appear to die of heat in the bottom 4x slot I think I'm going to buy a 26 TOPs Hailo 8 and do PCIE bifurcation with a 4x M2 slot card (I have a 16x PCIE 4 slot spare now). I need to move an NVME from an onboard slot as it shares lanes with the port that now houses my NIC, so a GPU/ARC isn't an option. Ah well, I quite like the idea of 26 TOPs for 2.5W.
Just some measuring after my kid goes to bed then I'll get it all ordered. Hopefully I can start configuring the software at the weekend.
This will be my first NPU so a question if I may; do I need to set a flag on the Hailo to ensure it lets me and my family survive the rise of Skynet?
Just some measuring after my kid goes to bed then I'll get it all ordered. Hopefully I can start configuring the software at the weekend.
This will be my first NPU so a question if I may; do I need to set a flag on the Hailo to ensure it lets me and my family survive the rise of Skynet?
Gassing Station | Computers, Gadgets & Stuff | Top of Page | What's New | My Stuff


