[Config Support]: simple motion detection

[sulliwane]

Describe the problem you are having

Hi there!
Thanks for that awesome Firgate. So:

• I have HA running on a very lightweight raspberry pie type of hardware
• I have don't have a Coral.
• I just want to use Frigate for local retention of motion recordings.
• I just want to record on motion.
• I don't want to run complex tensorflow image recognition, heavy on CPU.
• I have this camera which native resolution is 2304 x 1296

I got everything working, but I see the CPU usage is very high. Also I see "Person: 77%" which most probably implies some image recognition algo is being run.

Which config would you recommend to be as low CPU as possible, and just get very basic recording on motion? Or is my current config enough?

Any help would be much appreciated 🙏

Bonus question: would getting a Coral reduce the CPU load for my specific use case ?
Bonus question 2: how to push HA notification on motion with snapshot only when a specific HA state is being checked and true (away from home)?

Version

0.13.2-6476F8A

Frigate config file

mqtt:
  enabled: True
  host: core-mosquitto
  port: 1883
  user: "<redacted>"
  password: "<redacted>"

motion:
  # Optional: The threshold passed to cv2.threshold to determine if a pixel is different enough to be counted as motion. (default: shown below)
  # Increasing this value will make motion detection less sensitive and decreasing it will make motion detection more sensitive.
  # The value should be between 1 and 255.
  threshold: 30

record:
  enabled: True
  retain:
    days: 3
    mode: motion
  events:
    retain:
      default: 30
      mode: motion

snapshots:
  enabled: True
  retain:
    default: 10

cameras:
  porte: # <--- this will be changed to your actual camera later
    enabled: True
    ffmpeg:
      inputs:
        - path: rtsp://<redacted>:<redacted>@192.168.1.12/live0
          roles:
            - detect

Relevant log output

N/A

Frigate stats

No response

Operating system

HassOS

Install method

HassOS Addon

Coral version

CPU (no coral)

Any other information that may be helpful

No response

[sulliwane]

Note: I updated the config a bit and it seems CPU usage decreased. Still, happy to get your feedback on how to improve even more!

Note: my camera native/full resolution is 2304 x 1296.

Note 2: currently my recordings seem to be in HD only, as you can see:

but it should be 2304 x 1296 no? Maybe RTSP is only available in 1920 x 1080...

Newest config:

mqtt:
  enabled: True
  host: core-mosquitto
  port: 1883
  user: "<redacted>"
  password: "<redacted>"

motion:
  # Optional: The threshold passed to cv2.threshold to determine if a pixel is different enough to be counted as motion. (default: shown below)
  # Increasing this value will make motion detection less sensitive and decreasing it will make motion detection more sensitive.
  # The value should be between 1 and 255.
  threshold: 30

record:
  enabled: True
  retain:
    days: 3
    mode: motion
  events:
    retain:
      default: 30
      mode: motion

snapshots:
  enabled: True
  retain:
    default: 10

cameras:
  porte: # <--- this will be changed to your actual camera later
    enabled: True
    ffmpeg:
      inputs:
        - path: rtsp://<redacted>:<redacted>@192.168.1.12/live0
          roles:
            - detect
            - record
    detect:
      enabled: true
      width: 1152
      height: 648
      fps: 5
    record:
      enabled: true

Thanks a lot in advance!

[sulliwane]

thanks a lot @NickM-27 for the help. puting detect: false seems to have disable motion detection...

I don't see any new events now. Strange

[NickM-27]

that is expected, events are only for object detections. In the next version of frigate which has an entirely new UI, there will be a tab in the review page to review motion

[hawkeye217]

Events are generated from objects that were detected. If you disable detection, events will never be saved.

But with motion recording enabled, Frigate will retain recording segments with motion. You can review those from the "Recordings" button for your camera.

[sulliwane]

Got it. I re-enabled detection then 😄

And checked the cpu of my board, I see "Rockchip" being mentioned:

# lscpu
Architecture:                    aarch64
CPU op-mode(s):                  32-bit, 64-bit
Byte Order:                      Little Endian
CPU(s):                          4
On-line CPU(s) list:             0-3
Thread(s) per core:              1
Core(s) per socket:              4
Socket(s):                       1
NUMA node(s):                    1
Vendor ID:                       ARM
Model:                           4
Model name:                      Cortex-A53
Stepping:                        r0p4
CPU max MHz:                     1296.0000
CPU min MHz:                     408.0000
BogoMIPS:                        48.00
NUMA node0 CPU(s):               0-3
Vulnerability Itlb multihit:     Not affected
Vulnerability L1tf:              Not affected
Vulnerability Mds:               Not affected
Vulnerability Meltdown:          Not affected
Vulnerability Spec store bypass: Not affected
Vulnerability Spectre v1:        Mitigation; __user pointer sanitization
Vulnerability Spectre v2:        Not affected
Vulnerability Srbds:             Not affected
Vulnerability Tsx async abort:   Not affected
Flags:                           fp asimd evtstrm aes pmull sha1 sha2 crc32 cpuid
# uname -a
Linux homeassistant 5.15.35-rockchip64 #trunk SMP PREEMPT Sat Apr 23 12:10:09 UTC 2022 aarch64 GNU/Linux
# cat /proc/device-tree/compatible
rockchip,rk3318-boxrockchip,rk3328-boxrockchip,rk3328
#docker ps
ghcr.io/blakeblackshear/frigate:0.13.2

Looking at the Frigate Hardware Acceleration page, it seems I should use a -rk docker image, together with --privileged mode:

Any pointer on where the Dockerfile I need to edit is located? Also, do I need to run the "non protected" version of Frigate or not?

Finally, which docker image name is the correct/latest one? I saw a reference to this one ghcr.io/blakeblackshear/frigate:0.13.0-beta6-rk

Thanks a LOT for your time. Much appreciated 🙏

[NickM-27]

if you didn't setup a docker file then there is none

you can use stable-rk

[sulliwane]

sorry to look stupid, but where am I supposed to "use stable-rk"? I have SSH access to the board.

root@homeassistant:/usr/share/hassio# ls addons
core  data  git  local
root@homeassistant:/usr/share/hassio# ls addons/core/
assist_microphone  deconz	duckdns		  letsencrypt  matter_server  openthread_border_router	README.md     silabs_flasher	    tellstick  zwave_js
cec_scan	   dhcp_server	git_pull	  LICENSE      mosquitto      openwakeword		rpc_shutdown  silabs-multiprotocol  vlc
configurator	   dnsmasq	google_assistant  mariadb      nginx_proxy    piper			samba	      ssh		    whisper

[sulliwane]

ok my bad. Note: I'm now using a google coral for object detection.

In that case, I guess I should just go with the HA addon version of Frigate (which is the "stable" docker image, because the "stable-rk" won't bring anything on the table), and add this to the Frigate config:

ffmpeg:
  hwaccel_args: preset-rk-h264

Do I get it right? (and thanks a lot for the speedy reply 🫶)

Note:

root@homeassistant:~/services/frigate# uname -a
Linux homeassistant 6.3.13-rockchip64 #1 SMP PREEMPT Tue Jul 11 19:39:51 CEST 2023 aarch64 GNU/Linux

root@homeassistant:~/services/frigate# cat /proc/device-tree/compatible
rockchip,rk3318-boxrockchip,rk3328-boxrockchip,rk3328

[NickM-27]

I would suggest using docker

[sulliwane]

Hum, re-reading the documentation on hardware acceleration, I still need this:

but:

root@homeassistant:~/services/frigate# sudo cat /sys/kernel/debug/rknpu/version
cat: /sys/kernel/debug/rknpu/version: No such file or directory

root@homeassistant:~/services/frigate# uname -a
Linux homeassistant 6.3.13-rockchip64 #1 SMP PREEMPT Tue Jul 11 19:39:51 CEST 2023 aarch64 GNU/Linux

So I guess that adding hwaccel_args: preset-rk-h264 to the config will be useless

[sulliwane]

I would suggest using docker

May I ask what docker provides over HA addon (which is docker managed by HA supervised in my case)?

[NickM-27]

control over the system, like the ability to ensure the necessary files are provided to the docker container like are needed in this case

[saket424]

Is it possible to get a motion end event without incurring the rest of the overhead and CPU fluctuating due to object detection?

[NickM-27]

The new UI in 0.14 shows periods of motion

[saket424]

Is there a mqtt topic that can notify us of these periods of motion so we can get notified and do further post processing with some of these super duper large vision models and create manual events summarizing and annotating the recordings

[NickM-27]

Yes, see the MQTT docs

[saket424]

Thanks @NickM-27 . I understand better

https://docs.frigate.video/configuration/motion_detection/
https://docs.frigate.video/integrations/mqtt/