I am running Hypriot.
1 Like
Have you run benchmark.py on the Odroid-H2? I am curious what the inference speeds are.
Will this image work with a Raspberry 3?
I havenāt tried, but I think so.
Hi, Iām on RPI4 and when I run:
sudo docker run --rm --privileged -v /dev/bus/usb:/dev/bus/usb -v /home/pi/frigate/config:/config:ro -p 5000:5000 -e RTSP_PASSWORD='admin' blakeblackshear/frigate:0.2.2-beta
Iāve got this error:
Traceback (most recent call last):
File "detect_objects.py", line 99, in <module>
main()
File "detect_objects.py", line 53, in main
On connect called
cameras[name] = Camera(name, config, prepped_frame_queue, client, MQTT_TOPIC_PREFIX)
File "/opt/frigate/frigate/video.py", line 122, in __init__
self.rtsp_url = get_rtsp_url(self.config['rtsp'])
File "/opt/frigate/frigate/video.py", line 63, in get_rtsp_url
rtsp_config['path'])
KeyError: 'path'
Any help would be appreciatedā¦
Thanks
I was able to get hardware acceleration working on the RPi4 for decoding the h264 stream. Unfortunately, I discovered that there is no acceleration available on the ARM architecture for converting from yuv420p to rgb24 in ffmpeg. Because of this, the CPU usage isnāt much better with it enabled. 
Thatās a shame. Iām also running on an rpi 4 and got excited about your previous comment about hwaccel. Maybe I need to buy a usb3 pci card for my server. Appreciate the multi-arch container though! Was taking forever to build on my pi.
I finally got time to get on this great piece of software. Thank you @blakeblackshear for developing this!
Did someone get this running using Android IPWebcam as a rtsp source?
Iām running the odroid branch on an Odroid XU4 with the following config:
web_port: 5000
mqtt:
host: mosquitto
topic_prefix: frigate
user: frigate
password: frigate
cameras:
ipwebcamdoor:
rtsp:
user: ipwebcamdoor
host: <host>
port: 8080
password: password
path: /h264_ulaw.sdp
regions:
- size: 720
x_offset: 0
y_offset: 0
min_person_area: 5000
threshold: 0.5
And I get the following error:
frigate | On connect called
frigate | [h264_v4l2m2m @ 0x8095b0] capture: driver decode error
frigate | INFO: Initialized TensorFlow Lite runtime.
frigate | Creating a new capture process...
frigate | Starting a new capture process...
frigate | Capture process for ipwebcamdoor: 26
frigate | * Serving Flask app "detect_objects" (lazy loading)
frigate | * Environment: production
frigate | WARNING: This is a development server. Do not use it in a production deployment.
frigate | Use a production WSGI server instead.
frigate | * Debug mode: off
frigate | * Running on http://0.0.0.0:5000/ (Press CTRL+C to quit)
frigate | Opening the RTSP Url...
frigate | [h264_v4l2m2m @ 0x807990] capture: driver decode error
I did get it to work, I had VLC Player open on the same stream. Once I closed it frigate worked without problems.
Thanks a lot for the great job, it take me a while but finally works.
The only problem is that after some time the video is like the image below and some time it recognize my dog as a person.
Iām on Rpi 4.
Ok lowered the resolution do the job. Now it works like a charm.
thanks
is there a way to run this without docker? I donāt run docker or hass io
Technically the Dockerfile contains all the things you would need to install. Honestly, you should just run docker even if itās just for this. There are some heavy dependencies and upgrades are going to be a complete nightmare. Using docker will be infinitely easier than trying to reverse engineer and maintain your install manually.
2 Likes
Well after ordering my Coral back on the 4th of April!!! I have finally just got this setup, thanks for your hard work.
Just need to finish the automationās off.
Looking forward to see if you can get it to recognise previously seen people at some point in the future.
whatās up with your camera? Very hard to tell whatās going on in that scene.
Was the framerate too highā¦for rpi 4 1080 itās too much.
Could someone please share their frigate config.yml file for Reolink cameras? The feed I use is visible in HomeAssistant and VLC but i cant for the life of me get the Frigate generated camera to work.
Also, user and pass needed in the config file if its all contained in the streaming URL?
I dont have those cameras. I have unifi but here is an exampleā¦dont use the full url with un/pw included. break it apart.
HypriotOS/armv7: pirate@pirate in ~
$ cat /opt/frigate/config/config.yml
web_port: 5000
mqtt:
host: 10.1.99.113
topic_prefix: frigate
cameras:
unifi:
rtsp:
host: 10.2.2.2
port: 554
path: /live
user: this
password: could
regions:
- size: 350
x_offset: 0
y_offset: 100
min_person_area: 5000
threshold: 0.6
- size: 300
x_offset: 350
y_offset: 130
min_person_area: 5000
threshold: 0.6
# unifi2:
## rtsp:
# host: 10.2.2.2
# port: 554
# path: /live
# user: this
# password: could
# regions:
# - size: 400
# x_offset: 0
# y_offset: 100
## min_person_area: 5000
# threshold: 0.6
# - size: 300
# x_offset: 400
# y_offset: 200
# min_person_area: 5000
# threshold: 0.6
unifi3:
rtsp:
host: 10.1.1.2 ##IP ONLY
port: <YOUR_PORT_NUMBER>
path: /LIVE ###WHATEVER IS AFTER THE '/' RTSP://TEST:[email protected]:554/<THIS PART>
user: ubnt ##USER NAME FROM ORIGINAL URL
password: ##PW FROM ORIGINAL URL
regions:
- size: 330
x_offset: 0
y_offset: 75
min_person_area: 5000
threshold: 0.6
- size: 330
x_offset: 330
y_offset: 75
min_person_area: 5000
threshold: 0.6
# - size: 660
# x_offset: 1280
# y_offset: 400
# min_person_area: 5000
# threshold: 0.6
1 Like
How do I limit my alerts to a certain person recognition confidence factor (ie 90%+)? I have been using frigate for a few months now (currently running the latest version) and i often get alerts for low % probability persons. For example headlights will trigger my car parked on the street as being recognized 60% person or my mailbox will send a flood of 40% person alerts. Looking to limit alerts to > 90% confidence factor. Thanks!
Itās in your config.yml. Last line in region I believe. āthresholdā