Opinions? Should I add an NVR to my system? If so, how long to most out here “loop”? Or does everybody use only trigger based?
I played with ZoneMinder and Shinobi a long while back (both can integrate with OH, ZoneMinder through an add-on and Shinobi through REST API calls). In the end they were consuming way too many resources for the benefit I got so I cut them out. I use the Wyze app/Nest app for anything I need to do that an NVR would provide (which is almost nothing).
Check out Frigate.
Jim T. Star Fleet. Uh. Your last name doesn’t start with K does it?
LOL, no, and I wasn’t born in Iowa, nor on a star ship.
But seriously, I’ve just changed from Blueiris to Frigate after dragging my feet for so long. I quite like it so far:
I use it for:
- Announcing when a vehicle had just pulled into the driveway
- Announcing when and how many people were detected on the driveway or front porch. At night, this turns the outside lights on.
- Triggering my openhab rule that performs licence plate recognition
- And the primary reason is obviously to record and keep an eye on things. I can also access the recordings/events while I’m away by using VPN
I assume you do not continuously RECORD. But probably can (if desired) STREAM?
And… VPN. I actually like that idea instead of (again) a cloud solution. It is under my control.
It might be worth mentioning that NVRs often perform much better when given bare metal to run on with a good GPU. When I complained above that they were using too many resources, it was because I was running them on a VM without passthrough access to a real GPU so even keeping up with four streams with motion detection and recording took more CPU/RAM than I was willing to dedicate to it. YMMV.
Just something to consider for those looking to deploy an NVR.
I do continuously record to a dedicated 10TB drive, but you can configure Frigate to just record the events. My linux server is running on bare metal (not vm), and Frigate runs inside docker. I am using an nvidia gpu to help with de/encoding + inference.
I’m going to be running OH (when I’m done playing) on my Plex server. i5 of some speed, intel and nVidia graphics, and only 16GB memory. BUT… OH shouldn’t be an issue. I am LOOKING at possibly using it as my NVR too. But I guess the only way to know is to try.
That’s what all this Hobby stuff is about, right?
How about my old clunker Duo Core 2.7Ghz with 4GB memory, onboard shared memory graphics, and XP? That should work, right? LOL
Jim, what cameras are you using with Frigate? I have some cheap Wansview cameras and they really don’t work well.
Dahua of various models (4k, 5mp, 1080p). Their models are all similar in their rtsp capabilities. They offer full resolution for the main stream and a 720p substream for frigate’s detection and inferencing.
They also offer in-camera detection (movement, person, vehicle) and various rules like tripwire, direction detection, zones, etc. Although frigate can do some of what the camera can do, with a nicer mqtt interface.
Most likely not going to cut it. Frigate also needs Linux I think.
If the CPU supports Intel’s quicksync feature for the acceleration, or the nvidia card is supported by frigate it might be enough. I guess give it a try.
My gut feeling says you might need at least more ram which is probably the cheapest and easiest upgrade to do. Also which nvidia card?
So far I have been using doods. As there is a successor ( doods2 ) I also wanted to look into alternatives and read about frigate before it was raised here.
frigate/doods2 both support Edge TPU which I am going to look into.
Not necessarily. They only offer a Docker image so it should be able to run anywhere Docker can run a container. However, I believe Docker for Windows doesn’t support hardware passthrough (yet?) so if you plan on using it with Coral or GPU acceleration or the like you might need Linux.
Recommended hardware | Frigate has their recommended hardware.
Jim T - It was a joke. I can’t imagine that even coming CLOSE to being useful.
I’m running Blue Iris on a Windows VM (I give it 4 cores of a NUC 10 i7 and 8GB RAM) and pass through Intel graphics which makes a huge difference. I have 7 cameras, CPU usage is <10%, I offload people detection duties to codeproject ai server and notify openhab only when a person is detected.
GPU passthrough and substreams are essential to low resource usage.
I’m recording my outside cameras continuously and the internal cameras only during the night if the zwave motion sensor I have in the same room triggers.
I avoid Windows at all costs usually but have actually found Blue Iris in this configuration to be very stable.
i use orchid cord from ipconfigure in an ubuntu VM and i am really satisfied. I have tried many programms before (zoneminder, Shinobi and bluecherry).
I run my cameras through Camect and I wrote a simple app that connects to it and listens for events from the camera then sends them into OH via MQTT messages. I haven’t done much with it yet really but I can collect things like:
“person detected by front porch camera”
I was hoping to eventually drive some interesting welcome light shows via wled near holidays, etc.