Local realtime person detection for RTSP cameras

Look in the updated docs

Link to docs is in release docs

Thanks
I did look there, and couldnā€™t see it anywhere

Iā€™m looking for a better solution for my local NVR to replace an older copy of Blue Iris. I ran across Frigate and it sounds like just the ticketā€¦ right up until I get to the supported hardware only to find that essentially none of it is available for purchase. Every vendor is showing year+ lead times to get more Coral devices, so Iā€™m holding out little hope of getting my hands on one any time soon.

So hereā€™s a question - does Frigate make sense if you just have a normal ESXi farm without a TPU? How does it compare vs Shinobi etc when using a CPU (meaning, using the only way to run Frigate unless you bought a TPU last year)?

depends I guess. I would believe that if 24/7 recording is only desire then frigate can fill that just fine without TPU. Its the detection that eats CPU and without TPU I doubt you could do more than 4 camera at checking 1 frame every 5 seconds or so. If you have GPU you may be able to work something out.

I use shinobi and while it looked good it ran like crap in docker and really wasnā€™t much better than motioneye, which is what I used before frigate. I would use frigate over shinobi anyday. Shinobi has UI config which is nice but honestly I care about function and frigate has that.

1 Like

learned today that you can download portions of recording from Frigate URL. Thought i would post info here. hopefully this is added as UI

The URL already exists and works. http://frigate_IP:5000/api/camera_name/start/1638829800/end/1638830100/clip.mp4 The start and end times are in unix timestamps.

I used VLC to save provided network stream to file.mp4

1 Like

how many HD cams can you process with Intel*Ā® *Quick Sync on a newer CPU like IntelĀ® Coreā„¢ i7-1165G7? i would like to build an Ubuntu miniPC with this but the user has 16 HD cams. Will it be enough with this CPU for ffmpeg and the USB Coral for Deep Learning? @blakeblackshear Fanless Mini PC Core I7 with 8565U 8559U 8650U 10510U 10710U

What is the best way to filter out ā€œcarā€ from an area but keep recording all others? persons, cats, etc.?

I donā€™t know what country you are in but Farnell UK is showing stock for the Dual TPU version.
Although most motherboards / PCIE adaptors will only allow one TPU to work.

They are also due to get the USB one back in stock next month.

1 Like

Iā€™ve had my usb coral on order since september. Quite annoying.

But just so you know, if you have an igpu that does intel quicksync it works pretty well.

I have only a single camera fed into my HA which all runs on an asus chromebox (i3 4010u). Even this low end 4th gen processor can do hardware acceleration and I noticed a drop from 85% cpu utilisation to under 20% for my entire Home assistant installation just by enabling it in the frigate settings.

1 Like

Mouser here in the states is still showing 40+ week lead time butā€¦ itā€™s also the only shop allowing me to place a backorder so Iā€™ve gone ahead and did so. Maybe Iā€™ll be surprised, or maybe Iā€™ll get one in 2023. Either way, thanks for the tip!

I ordered from mouser, though not from the states. it sad October shipping when I ordered, still waiting.

did you try something like

objects:
 track:
   - person
#   - car
   - bicycle
   - motorcycle
   - bird
   - cat
   - dog

I have an issue where persons are being detected on the sidewalk in front of our house. I have created a motion mask and a object mask for person that covers the sidewalk. Still this morning I got a trigger on a person on the sidewalk.

What am I missing here?

I believe frigate uses the bottom of the bounding box to determine ā€œwhereā€ it is. Ensure the mask is far enough down the image is my only thought.

yes, but I want cars to be tracked, but not an areaā€¦

Is there in some way possible to stream a Frigate camera to a webb page?

Use the RTMP stream

But RTMP streams have to be converted when streaming to a webpageā€¦

This looks promising and installs in docker container

1 Like

Did try this a while agoā€¦ But in a another purposeā€¦ Will check this aganā€¦ :slight_smile: