Comment by 9x39
1 day ago
With face detection? License plates? Tamper protection?
I'm guessing you're thinking Reolink or other Chinese ultra-commodity cam. It's fine, it's just in a different product class and ecosystem - and that's where enterprises fit in, they want that support+ecosystem and not DIYing.
Reolink CX820 8MP $129 https://reolink.com/product/cx820/
Unifi G6 8MP ~$300 https://techspecs.ui.com/unifi/physical-security/uvc-g6-dome...
Avigilon H6A 8MP ~$1200 https://www.avigilon.com/security-cameras/h6a-dome
> With face detection? License plates? Tamper protection?
I do that with my Unifi Protect doorbell. RTSP streams. Google Coral. Frigate. Scales very well. Do ML on low quality stream. Look/save the high quality stream. You do it all centralized, and you can put the camera(s) on a seperate VLAN. They don't even need internet access. If you run them over PoE twisted pair, the attacker would need physical access to perform MITM. Wireless, one should assume the camera is insecure (e.g. KRACK).
Wow, that's cool, learned something new today. Does that work better in your estimation than the UI Protect software?
The purpose of my comment had only been pointing out those features don't come onboard a $100 cam.
(not the parent poster, but same setup): Is it better than UI Protect? No, but you can make it about the same.
I have the same popular setup (Frigate) although I just use ONNX on an 11th-gen Intel CPU instead of a Coral (unless you are trying to do something fundamentally goofy like use a Raspberry Pi as an NVR, Coral doesn't really perform better than even a several-generations-old iGPU or iNPU).
This is the typical OSS story: you can duct tape a giant leaning tower of janky stuff (Frigate + go2rtc + HomeAssistant + various connectors + some kind of VPN/proxy solution for away-from-home access) together and get something that's fairly close to the commercial solution, where you click a button. The open source solution is fun and more customizable in highly niche ways (you can bring your own image recognition models and tagging, adjust the resolution and encoding for everything in infinite detail, and so on) and the commercial solution is easy and works. Chose your path.
I will say I've liked the Frigate stack, though. I'm making some recognition tweaks for recognizing animals on my property, the software works well enough, and I do like having a really, truly on-prem solution for this specific thing.
You already got a good reply, but I can maybe add n+1 and some details:
It works similar, but requires some effort to get working (if you already self-host its peanuts think Frigate plus reverse proxy and I also use Wireguard to have it available from outside). My home connection is fiber 1 gbit, but with DSL (only 30 mbit upload) it worked fine, too.
Since I want to decrease my reliance on US cloud, I like to self-host. I also still rely on Unifi APs and the doorbell. Right now I wouldn't spend money on building a self-hosted server, given prices.
I should mention I use iGPU via SR-IOV on a VM. The Google Coral sits in the device unused.
I also immediately copy the stream to an offsite backup. This way, if I get coerced to destroy my doorbell feed, I will happily oblige.
I have rather a lot of Reolinks ... and Frigate on Home Assistant. The cameras are on a VLAN with rather minimal internet access (ie none) I make pool.ntp.org etc resolve to my own NTP servers too.
I never really thought of Ubiquity as enterprise always felt more of the premium small to mid sized business but I am sure some enterprises use them.
They definitely started in the turnkey SME/consumer space, and still do a lot that's relevant there, but they've got a ton of very enterprise hardware targeted at large offices/campuses/stadiums and the like. There's APs specifically designed to be pointed at a sports or concert crowd and handle 1500+ active clients per access point at a time, and similarly specced switches and routers so that you can have TV crews turn up and hook into your network for live streaming.
The new enterprise NVRs work pretty well.
I think they're definitely not Avigilon, Genetec, Verkada, but we run a few hundred UI cams in some edge areas. It works, esp if you don't demand orchestration.