Aha! Thanks, yes, it had to do with the min_person_area: parameter.
By the way, how does that parameter work? (I smell another clarifying update of the config.yml example file coming 
)
The person area is calculated by width x height of the bounding box. Anything smaller than the min_person_area is ignored for that region. It is designed to filter false positives when you know a person couldnāt be that small.
1 Like
Got my Coral and set this up on a Synology NAS over the weekend, itās working well.
Thanks for your efforts @blakeblackshear , lots of interesting things here.
I wanted to see what the debug data generated but enabling it took up so much CPU that the docker container became unresponsive. This is likely due to my setup though, so iāll mess with it some more. I believe my camera frame rate is too high (12 currently)
Are there plans to output other objects apart from People in the short term? as an example, i note that cars are detected in the preview stream. Would be great to save each best_car as well as an endpoint + dump all captured images to disk for review
2 Likes
Adding other object types is on the todo list. I am hoping to implement object tracking with dynamic regions first.
So I have hassio running on a NUC. Can I run this on a separate Raspberry Pi with a Google USB accelerator?
You could. The raspberry pi is about 10x slower because it doesnāt have usb 3.0. I am now running this on an Odroid-XU4 that is separate from my Hass server.
Why not run it on the NUC ?
I am also running HassIO on a NUC, installed portainer to manage docker.
After some fiddling I managed to get frigate running.
Because Iām a noob and did not install my hassio on docker or any VM 
I got an error message when trying to run the build command.
Step 18/20 : ADD frigate frigate/
ADD failed: stat /var/lib/docker/tmp/docker-builder155260706/frigate: no such file or directory
Do you know what the issue is or how I can fix it?
Edit: Tried rerunning the build command and got this error at the start, it continues to run, but didnāt see this error last time.
ERRO[0524] Canāt add file /home/tc23/.gnupg/S.gpg-agent to tar: archive/tar: sockets not supported
What is the build command you are running? Are you running it from the root of the source code? What OS?
I downloaded the GitHub and extracted the Dockerfile and placed in my home directory, /home/tc23/Dockerfile. From there I ran the docker build -t frigate . command. This is my first time using docker, so itās likely Iām missing something easy. Iām on Ubuntu 16.04, and installed docker-ce, per the official installation documentation. I am trying to install on a NUC
Thatās your problem. You need the entire directory from github, and you have to run the command from inside the directory. If you move the dockerfile out, itās not going to work.
That did it! I knew I was doing something stupid. Thanks!
I just wanted to pop in and say thank you again. Do you have a buy me a coffee link somewhere?
1 Like
Not at the moment. I may add something like that to the Readme in the future.
2 Likes
Have you made any progress on the hardware acceleration of stream decoding? I have been trying the to accomplish it for a similar project but without success
I already have it working for the next version. If you can run ffmpeg in a container on your machine with hwaccel, then you can use that same command with frigate. OpenCV doesnāt support it, so you have to run ffmpeg in a subprocess and read the raw video stream from stdout. That allows you to pass any params you want to ffmpeg. As a side benefit, you donāt have to convert BGR to RGB either.
1 Like
sounds great! Looking forward to the next version.
Brilliant! Do you have it on any of the dev branches so I can check it out?
Are you now running all your 6 cameras on an Odroid-XU4 instead of your i3 NUC and if so are you getting better performance?