Local realtime person detection for RTSP cameras

Walt made an incredible node-red automation for the native iPhone HA app here:

I’ve tried to recreate this for Android but I’m not having any luck. Has anyone created a flow for node-red and Android app with some advanced functionality? I’d really like to do two things, be able to mute notifications for X minutes off the notification and disable notifications if a door has been recently opened (don’t notify drive way camera for 10 minutes if the front door was opened).

I’m just not there with my programming know-how even if node-red is doing most of it. :smiley:

If you haven’t already, attempt to flash back to Annke firmware. My C800 has a dropdown to choose between h264 and h265.
I’m surprised Hikvision doesn’t have this same option. Try latest firmware maybe?
Best of luck! :crossed_fingers:

Hi,

thank you @blakeblackshear! It’s great tool :slight_smile:

I’m looking for install it on a jetson nano 2GB (for me google coral little bit expensive and my HA is running on intel nuc virtualbox). As I see in the forum it is possible. Can Frigate work on a jetson nano 2gb? And how can I install it on jetson nano, e.g. ubuntu + Frigate in docker?

Than you in advance!

Well done to you,… I was worried the naysayers would throw you off… Thanks for contributing back to us

1 Like

Thx m8 :slight_smile: :heart:

Look this guy use it

If you know linux, partly you can use this tutorial, but only part as its for intel integrated GPU
https://community.home-assistant.io/t/network-share-on-frigate-hassos-with-nabucasa-ssl-and-proxmox-lxc-container/
I’m now also interested in this jetson. If/when i buy i might make a tutorial.

Do somebody use Frigate on Raspberry pi 4? How is the performance?
Because i read this

TensorFlow Lite on Raspberry Pi 4 can achieve performance comparable to NVIDIA’s Jetson Nano at a fraction of the cost.

https://towardsdatascience.com/3-ways-to-install-tensorflow-2-on-raspberry-pi-fe1fa2da9104

works with two 720p streams at 5 fps detection, inference speed is 200-300ms

Thank you for reply, can you explain what is inference speed?
Do you think it can handle any more cameras as 5fps?

@blakeblackshear Hi, please can you explain why whenever i stop frigate the container just disapear from the listing in portainer.
P.S. cache is cleared. Other containers for example ssh are in stop state when i stop them from HASS GUI.
Thank you

You are probably running with --rm.

Thank you for reply, i’m using frigate addon. And i’m trying to figure out how to resetup the addon to make additional changes when the addon is started.
Is the --rm in your default setup/image?
Can it be removed?

H all and @blakeblackshear

I note a few people are having problems with filters and have read some of the fixes. But is anyone have trouble with a mask in a filter?

I have this in global config:

  track:
    - car
    - person
    - bicycle

And this under my front camera:

    objects:
      track:
        - car
        - person
        - bicycle
      mask:
        - 0,174,331,0,0,0

However the mask is ignored?

At the camera level, mask should be under ‘motion:’, or specified as part of the object filter - example:

    motion:
      delta_alpha: 0.6
      threshold: 18
      mask:
        - 1463,946,1508,1061,1286,1161,0,1809,0,0,1006,0,1767,0,1732,184,1640,178,1594,181,1548,204,1509,206,1462,247,1424,276,1459,421,1492,635
        - 3072,0,2491,603,2385,416,2305,327,2276,316,2178,376,2082,308,1965,242,2070,0
    objects:
      track:
        - person
        - car
        - dog       
      filters:
        car: 
          mask: 
            - 1042,942,1506,929,1517,338,1159,325       
        dog:
          mask:
            - 0,1779,0,2048,2919,2048,1623,1057,1466,1059
        person:
          mask:
            - 0,1779,0,2048,2919,2048,1623,1057,1466,1059

EDIT: Just RTFM-ed myself… seems you’re correct, mask can also apply directly below object:

… did you try it as per the example, just in case?

objects:
  track:
    - person
    - car
  mask: 0,0,1000,0,1000,200,0,200

Looks like Google Coral USB TPUs are starting to come into stock, my order from Mouser I put in back in the beginning of March just shipped. Hopefully others will see them soon too.

I have managed to mount a network share to the host system HASSOS to allow frigate to save to network share
Tutorial:
https://community.home-assistant.io/t/solved-hassos-mount-nas-network-share/

Other tutorial that i did before with proxmox container is depreciated

1 Like

I’ve setup an automation via nodeRED to turn off detection of my driveway camera for a few minutes when our front door is opened. I have zigbee door sensors so use those to turn off the Frigate detection for 3 minutes, then turn detection back on. It obviously doesn’t record a clip of any detections, so not exactly what you asked for but it works for my use case.

I’m a newbie to Node-Red, and have tried a few different examples for android notifications but have never got it working correctly. I use a yaml automation for android notifications which works ok, but would rather keep everything in NodeRED.

Ive done something similar in yaml with 2 automations.

- id: '1612466944061'
  alias: Camera de-activate
  description: Deactivate front camera triggers on front door opening
  trigger:
  - platform: state
    entity_id: binary_sensor.front_door
    from: 'off'
    to: 'on'
  condition: []
  action:
  - service: switch.turn_off
    data: {}
    entity_id: group.camera_disable
  mode: single

- id: '1612467240719'
  alias: Camera re-activate
  description: Deactivate front camera triggers on front door opening
  trigger:
  - platform: state
    entity_id: binary_sensor.front_door
    from: 'on'
    to: 'off'
  condition: []
  action:
  - delay:
      hours: 0
      minutes: 15
      seconds: 0
      milliseconds: 0
  - service: switch.turn_on
    data: {}
    entity_id: group.camera_disable
  mode: single

and grouped the camera’s I’m using

Camera_disable:
  name: Camera disable
  icon: mdi:cctv
  entities:
    - switch.doorcam_detect
    - switch.drivewaycam_detect

It works well enough for us.

Any sugestions on settings?
when i remove this,which is recomended in docs it is working.

ffmpeg:
  hwaccel_args:
    - -hwaccel
    - vaapi
    - -hwaccel_device
    - /dev/dri/renderD128
    - -hwaccel_output_format
    - yuv420p

I’ve got green screens on all 12 cameras,stream is H264
But i want to enable the passthrough integrated GPU that i pass to be operational.
Intel <10gen settings

Thank you.
If anybody have NUCs without corral, maybe your settings should work? please let me know.

frigate.video                  INFO    : CAM4: ffmpeg process is not running. exiting capture thread...
frigate.video                  INFO    : CAM3: ffmpeg sent a broken frame. memoryview assignment: lvalue and rvalue have different structures
frigate.video                  INFO    : CAM5: ffmpeg sent a broken frame. memoryview assignment: lvalue and rvalue have different structures
frigate.video                  INFO    : CAM3: ffmpeg process is not running. exiting capture thread...
frigate.video                  INFO    : CAM2: ffmpeg sent a broken frame. memoryview assignment: lvalue and rvalue have different structures
frigate.video                  INFO    : CAM2: ffmpeg process is not running. exiting capture thread...
frigate.video                  INFO    : CAM5: ffmpeg process is not running. exiting capture thread...
frigate.video                  INFO    : CAM1: ffmpeg sent a broken frame. memoryview assignment: lvalue and rvalue have different structures
frigate.video                  INFO    : CAM1: ffmpeg process is not running. exiting capture thread...
frigate.video                  INFO    : CAM7: ffmpeg sent a broken frame. memoryview assignment: lvalue and rvalue have 
ffmpeg.CAM_technicka.detect    ERROR   : Consider increasing the value for the 'analyzeduration' and 'probesize' options
ffmpeg.CAM_technicka.detect    ERROR   : [h264_v4l2m2m @ 0x55d40a815800] Could not find a valid device
ffmpeg.CAM_technicka.detect    ERROR   : [h264_v4l2m2m @ 0x55d40a815800] can't configure decoder
ffmpeg.CAM_technicka.detect    ERROR   : Error while opening decoder for input stream #0:0 : Invalid argument
ffmpeg.CAM8.detect             ERROR   : [h264_v4l2m2m @ 0x55595fafa580] Could not find a valid device
ffmpeg.CAM8.detect             ERROR   : [h264_v4l2m2m @ 0x55595fafa580] can't configure decoder

i don’t know how to try this suggestion?!

Okay so this is working:

Is this the prefered solution for integrated GPU?

I am unable to start the add-on. I get the following error is the Supervisor Log:

21-04-28 07:20:43 ERROR (SyncWorker_5) [supervisor.docker] Can’t start addon_ccab4aaf_frigate: 500 Server Error for http+docker://localhost/v1.41/containers/dd9eacd7ee8db5eee7889870ab035d7307e7185b3eda88652b2c2aaaf31160a5/start: Internal Server Error ("driver failed programming external connectivity on endpoint addon_ccab4aaf_frigate (09d24995e8e6a564e6453d9a6efe77eed595814c38fce80807