Local realtime person detection for RTSP cameras

Vendo232 · May 7, 2021, 7:46pm

Neofetch is the name of the app

ukro · May 7, 2021, 7:57pm

can you point me to some explanation what is inference?

etch-a-sketch · May 7, 2021, 8:14pm

Can anyone offer advice to get the USB TPU working on a synology DiskStation when utilizing docker? I am a real linux novice so I am not sure if the diskstation has ‘drivers’ for it and what I would need to do to pass it through to the docker instance.

ukro · May 7, 2021, 8:19pm

If the diskstation is under linux,no drivers should be needed i think

mr6880 · May 8, 2021, 12:15am

What do you see here on your debug page? I also have 11 cameras running.

tmjpugh · May 8, 2021, 12:31am

I installed the runtime software and udev rules on host mentioned in “setup” page at coral.ai

danbutter · May 8, 2021, 12:49am

From the docs:
Time spent running object detection in milliseconds.
/
“inference_speed”: 10.48,
/

So basically how long until an object is identified by mobilenet or whatever frigate is using for image processing/ object detection these days.

etch-a-sketch · May 8, 2021, 1:05am

I tried to follow the guide and got message any suggestions:

tee: /etc/apt/sources.list.d/coral-edgetpu.list: No such file or directory
deb https://packages.cloud.google.com/apt coral-edgetpu-stable main

EIDT: got this done by making the directory first. which i needed to do using ‘sudo mkdir’

tmjpugh · May 8, 2021, 1:12am

Did you add coral library to apt sources list?

echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list

curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -

sudo apt-get update

EDIT
Since it is synology I’m not sure if this apply since I know some things don’t follow Linux way exactly

etch-a-sketch · May 8, 2021, 1:25am

it turns out synology does not have ‘apt’

Vendo232 · May 8, 2021, 3:26am

Im not sure but Inference has something to do with picture processing, if you have Coral the inference could be below 10ms

Raspberry pi 4 - im getting 180ms with 3 full hd cameras without hw acceleration

T420 HP with N4000 and 600 GPU - 1200ms with the same 3 cameras.

Both without Coral

For some reason the Raspberry pi 4 appears to be working better than the HP Thin Client.

Vendo232 · May 8, 2021, 4:28am

I have t420 thin Client which comes with M.2 wifi this one. Would anyone be able to confirm if Coral M.2 A+E work?

Wireless-AC Intel 9260 9260NGW Adapter For M.2 Key E NGFF Wifi Network Card 9260AC, Wi-Fi + Bluetooth 5.0, Up to 1730Mbps-5G, 300Mbps-2.4G, Dual Band, IEEE 802.11ac, MU-MIMO, Windows 10 for Laptop

ukro · May 8, 2021, 7:31am

Screenshot_1121242×288 2.16 KB

ukro · May 8, 2021, 7:33am

Thank you,now i get it

Murph24 · May 8, 2021, 7:50am

I’m also running frigate in a docker container in a proxmox LXC. See if this helps. I could not get QSV working on my 10th gen intel, but vaapi will work. My CPU idles at ~9%. Also note I did a lot of messing around, so some of these settings may not be doing anything useful.

Proxmox container config (file 105.conf in my case). lxc.cgroup and lxc.mount are used to pass GPU device and USB for a coral stick"

arch: amd64
cores: 2  
features: nesting=1 
hostname: Docker 
memory: 4096 
mp0: /shared,mp=/shared,size=0 
net0: name=eth0,bridge=vmbr0,firewall=1,hwaddr=00:00:00:00:00:00,ip=dhcp,type=veth 
onboot: 1 
ostype: debian 
rootfs: local-lvm:vm-105-disk-0,size=115G 
startup: order=2 
swap: 512
lxc.cgroup.devices.allow: c 226:0 rwm
lxc.cgroup.devices.allow: c 226:128 rwm
lxc.cgroup.devices.allow: c 29:0 rwm
lxc.cgroup.devices.allow: c 189:* rwm
lxc.apparmor.profile: unconfined
lxc.cgroup.devices.allow: a
lxc.mount.entry: /dev/dri/renderD128 dev/dri/renderD128 none bind,optional,create=file 0, 0
lxc.mount.entry: /dev/bus/usb/002/ dev/bus/usb/002/ none bind,optional,create=dir 0, 0
lxc.cap.drop:

frigate config:

hwaccel_args: 
   #Optional: global input args (default: shown below)
    - -hwaccel
    - vaapi
    - -hwaccel_device
    - /dev/dri/renderD128

Docker run command passing USB and GPU devices:

docker run --name frigate --privileged --shm-size=1g --mount type=tmpfs,target=/tmp/cache,tmpfs-size=1000000000 -v /shared/frigate/config:/config:ro -v /shared/frigate/clips:/clips:rw -v /etc/localtime:/etc/localtime:ro -v /dev/bus/usb:/dev/bus/usb --device-cgroup-rule="c 189:* rmw" --device=/dev/dri/renderD128 -d -p 5000:5000 -e FRIGATE_RTSP_PASSWORD='password' blakeblackshear/frigate:0.8.4-amd64

Also, a useful container to see if your hardware accel is working:
https://hub.docker.com/r/djaydev/intel-gpu-tools

sender · May 8, 2021, 8:47am

Which CT Template did you use?

wills106 · May 8, 2021, 10:05am

You don’t need Coral drivers for the USB version if you are using Docker. But on the Synology you either need to Start the container from the command line so you can pass through the USB or take a look at Using Google Coral on Synology Docker without starting from CLI · Discussion #364 · blakeblackshear/frigate · GitHub

etch-a-sketch · May 8, 2021, 3:09pm

This did the trick. Took a little bit of interpretation but it worked. Thank you.

I wonder if frigate is ok for the hard drives. I know it quite intensive.

wills106 · May 8, 2021, 3:29pm

I wouldn’t exactly call Frigate intensive?
It’s going to be recording a lot less than a normal NVR as it’s only recording clips of actual people (or whatever trackers you have set). With a normal NVR it will be recording every time the sun moves or a leaf blows on a tree.

Murph24 · May 8, 2021, 3:49pm

Debian Turnkey Core I believe. Watch the first part of this where he sets up the LXC and docker. You don’t need any of the NGNIX stuff he installs towards the end