Local realtime person detection for RTSP cameras

In my previous automation I exposed the mp4 video generated by Frigate using another nginx container, sharing volume with frigate, on a public port. The video was not in the notification: just a link to launch the browser or the VLC player. Could you just expose the mp4 file using a similar api endpoint?

Should be easy enough. Can you open an issue on GH?

Done: https://github.com/blakeblackshear/frigate-hass-integration/issues/4

Another question: the documentation doesnā€™t say anything about the (old but still present) false_positive field in events and the notification examples report both kind of events.

Is something changed in its management? I remember that any event started as false_positive=true and, upon confirmation, it switched on false_positive=false and never changed.

The messages sent to frigate/events donā€™t start until false_positive is false in this version. The first publish to that topic for an event will be when false_positive transitions from true to false.

is this M.2 Accelerator something you can plug into an NUC to give the same functionality of the Coral USB device? https://www.mouser.co.uk/new/google-coral/coral-M2-accelerator-dual-tpu/
If so is it better/worse performance? (I see itā€™s a fraction of the price!) Is there a list somewhere that tells you which devices Frigate can use as accelerators?
What about this thing that can apparently run Debian Linux & has Edge TPU built inā€¦ suitable for Frigate? https://www.mouser.co.uk/new/google-coral/edge-tpu-dev-board/
thanks!

Any Coral device that works according to the documentation on coral.ai should work with Frigate. The M.2 Accelerator should work, but requires a different driver that isnā€™t included in HassOS. It isnā€™t compatible with every slot type. Performance should be the same and it also includes throttling if the temperature gets too high. The USB is the most compatible because it just presents as a raw USB device and the drivers are included in Frigate.

Personally, I would stay away from the dev board, but I havenā€™t tried it. It will probably be more complicated to leverage the GPU for ffmpeg decoding than using a RPi which has a much larger community.

I would love to find a minipc device or SBC with a built in pci express slot. The new asus tinkerboard looks promising: https://www.tomshardware.com/news/tinkerboard-2-and-tinkerboard-2s-announced

cool thank you!!

Loving the new 0.8.0 Seems like itā€™s all starting to pull together quite nicely!
Especially with the new integration.

Another ongoing issue during transition of automation: is there a public api to get the full image corresponding to the cropped thumbnail available using the public endpoint `ā€˜https://your.public.hass.address.com/api/frigate/notifications/{{trigger.payload_json[ā€œafterā€][ā€œidā€]}}.jpgā€™ ?

In order to inspect the event by mean of notification, I was used to make the click on thumbnail to open (on the browser) the full image. This would be a complement to the link to open the video. Should I add this request of the just opened FR on GH?

I would create a separate issue. The mp4 will be easy since it already exists. The full image is not stored anywhere currently.

Done: https://github.com/blakeblackshear/frigate-hass-integration/issues/5

Just a quick question, whatā€™s the main difference between:
save_clips:
record:

From what I am thinking?
save_clips: is when Frigate has detected a track model such as person

record: is when ffmpeg detects motion and outputs it regardless if Frigate detects a model or not? So more handy for debugging detection sizes. Or if you want a recording regardless if there is something there or not.

Recording Clips (aka save_clips)
24/7 Recordings (aka record)

I should probably rename those headings to make it more clear.

I have enough 24/7 recordings now to see it literally is 24/7 recordings and not motion based.

Love how little CPU is used on 0.8.0 compared to 0.7.3 on my Synology.

First time user. Installed Debian on a NUC and then Docker and Docker Compose. Iā€™m using the Docker compose file listed in the readme:

frigate:
    container_name: frigate
    restart: unless-stopped
    privileged: true
    image: blakeblackshear/frigate:stable-amd64
    volumes:
      - /dev/bus/usb:/dev/bus/usb
      - /etc/localtime:/etc/localtime:ro
      - <path_to_config>:/config
      - <path_to_directory_for_clips>:/clips      
      - type: tmpfs # 1GB of memory, reduces SSD/SD Card wear
        target: /cache
        tmpfs:
          size: 100000000
    ports:
      - "5000:5000"
    environment:
      FRIGATE_RTSP_PASSWORD: "password"
    healthcheck:
      test: ["CMD", "wget" , "-q", "-O-", "http://localhost:5000"]
      interval: 30s
      timeout: 10s
      retries: 5
      start_period: 3m

However Iā€™m getting the following error when running docker-compose up on that yaml configuration.

ERROR: The Compose file './docker-compose.yml' is invalid because:
Unsupported config option for frigate: 'healthcheck'
frigate.volumes contains an invalid type, it should be a string

Iā€™m new to both Docker and Docker-Compose so I could be doing something wrong.

Ideas?

Check your docker-compose version and the version of the docker compose file.

Stopping by to say thank you for the beta! I have a few Wyze and an Arlo that my wife suggested to switch over to Wyze, but I wasnā€™t happy with all of the false notifications in the summer. We mostly use it to see who is at our door and what animals are around our house. I started playing with the RTSP firmware (not the new one) on an extra cam I use for my smoker to monitor temps, and got setup with motioneye then Frigate. I got it setup and had person detection on it, but it would never be wife approved. This is getting SO MUCH closer to easy to use regarding clips and such! Not to mention the integration is awesome!!

Iā€™ll still be playing with it for a bit before I start loading RTSP on all my cameras and replacing the Arlo, but thank you very much for all of your work!

Question: Iā€™m curious about the models/labelmap - are there more models available? Is there a list that I can change the labelmap to? Curious for just generally an ā€œanimalā€ label or if I can track the raccoons that love to jump my fence every nightā€¦

Iā€™m not that experienced with all this yaml stuff, so I probably making some basic mistake.
Iā€™m trying to get the automation work in 0.80 beta using the provided example (but changed to telegram). But it fails to validate the code:

- alias: "When a person enters a zone named oprit"
  trigger:
    platform: mqtt
    topic: frigate/events
  conditions: 
    - "{{ trigger.payload_json["after"]["label"] == 'person' }}"
    - "{{ 'oprit' in trigger.payload_json["after"]["entered_zones"] }}"
  action:
    - service: notify.telegram_yj
      data_template:
        message: "A {{trigger.payload_json["after"]["label"]}} has entered the oprit."
        data:
          photo:
            - url: 'https://hass.valid_dns_name.nl/api/frigate/notifications/{{trigger.payload_json["after"]["id"]}}.jpg'
              caption: '{{trigger.payload_json["after"]["id"]}}
Error loading  while parsing a block collection
in "refers to the 1st condition line" column 5
expected <block end>, but found '<scalar>'
in "refers to the same condition line" column 33.

Any hints?

You can rename several entities to ā€œanimalā€ as described here: https://github.com/blakeblackshear/frigate/tree/release-0.8.0#customizing-the-labelmap

The models/labelmaps are provided by google, but frigate only works with a very specific type of model, so there arenā€™t any other out of the box options I am aware of.

1 Like