Some cameras use different authentication and may not allow the method of putting the username:password into the url. Try to dig up the specs for whatever firmware your camera is on.
The url with username and password works in VLC.
Could be a strange character in the password
Password is the same of the other working model. Also tried to change to a different one but no luck.
remove $ in the password if you have, I think I had same problem.
Dumb question perhaps but I did a search with no results.
Is there a gui to handle cams/settings etc?
Thinking about run this instead of Zoneminder.
I was hoping to look into creating one for some of this. Not on the roadmap to change settings yet though.
https://github.com/blakeblackshear/frigate/discussions/479#discussioncomment-263560
I have a strange issue.
I’m looking at a snapshot of the mqtt cameras made with the HA integration and running the .80 beta 3. I have found with one of my two cameras and only when it detects cars with that one camera…the entire image is black and white. This includes the bounding box. The same camera shows a normal color picture for the snapshot when it detects people. The clips made by this camera when detecting cars are full color. Only the snapshot is black and white. Anybody else seen this?
I have saw some mentions on GitHub of increasing gpu_mem on a Pi4 (4GB) - is this still necessary/advisable?
Moreover, I am having real issues with two 480p sub streams on a Pi4 with HassOS & custom_component. The video feed would be all green or 90% full of artefacts. When looking at the HTTP endpoint there would be multiple bounding boxes rapidly appearing and disappearing. It really was not stable.
I increased the tmpfs size to 512m and it seems to resolve it if anyone else is having similar issues.
1 Like
Sorry, how you change tmpfs size in the addon? Thanks
save_clips:
max_seconds: 300
tmpfs_cache_size: 512m
retain:
default: 10
objects:
person: 15
1 Like
Any pointers to how I would do this line on a supervised install on Debian
Note: You also need to set LIBVA_DRIVER_NAME=radeonsi as an environment variable on the container.
I’m not very proficient in docker - I installed it through the supervisor. I can find some of the docker JSON files but not sure where I would add this line. I am running this on an HP T620 Thin client which has Radeon HD 8xxxx integrated graphics.
Many thx, i got it working now!
Is there a way to change the default clips folder in the 0.8.0 addon? Cant find the option in the docs just for the old config.
Is anyone else having issues when restarting Frigate Beta Addon when HassOS is running. When i do that the above issue occurs.
To fix it, I need to stop the addon, restart home assistant and then start addon again. very annoying
I had similar issues on my reolink cameras when using the RTSP URL, another report indicated switching to the RTMP URL resolved it and it did for me as well. You might see if you have alternate protocols you can try. I don’t think it was a HaSS or Frigate issue I think that ffmpeg is just picky about glitches in the stream
If it helps, after some trial and error this is the setup that I got working with my Reolink RLC-410-5MP cameras. Initially with the older non-beta build, the RTSP streams were working fine, but when I switched to the 0.8 beta they had a lot of smearing. I switched to RTMP which improved things, but then when I tried to record clips off the RTMP stream the camera crashed of froze resulting in an FFMPEG crash.
Solution: Use the RTMP stream for detection, and the RTSP stream for clips:
detectors:
coral:
type: edgetpu
device: 'usb:0'
mqtt:
host: 192.168.86.xx
port: 1883
user: user
password: password
ffmpeg:
input_args:
- '-avoid_negative_ts'
- make_zero
- '-fflags'
- nobuffer
- '-flags'
- low_delay
- '-strict'
- experimental
- '-fflags'
- +genpts+discardcorrupt
- '-use_wallclock_as_timestamps'
- '1'
cameras:
porch:
ffmpeg:
inputs:
- path: >-
rtmp://192.168.86.xx/bcs/channel0_sub.bcs?channel=0&stream=0&user=user&password=password
roles:
- detect
- path: 'rtsp://user:[email protected]:554//h264Preview_01_main'
roles:
- clips
height: 480
width: 640
fps: 5
save_clips:
enabled: true
pre_capture: 5
post_capture: 5
objects:
- person
retain:
default: 7
objects:
person: 7
rtmp:
enabled: false
snapshots:
show_timestamp: false
draw_zones: false
draw_bounding_boxes: true
crop_to_region: false
height: 480
objects:
track:
- person
save_clips:
max_seconds: 60
retain:
default: 7
objects:
person: 7
This works well, with good detection and saving full-res clips.
It does seem to use more CPU than the older version (~60% vs. 25-30%), and for some reason I simply can’t get my NAS drive to mount using a shell command as described in the thread above, but works fine with local storage.
2 Likes
Interesting I have a 410 as well and use the RTMP stream for both without issues, tons of clips…I have the same parameters in ffmpeg and running at the same resolution/fps. I am on all wired… do you have wireless in the mix? Is your camera firmware up to date?
Latest firmware, POE wired. It may be because I was trying to record from the full-res stream? Or perhaps because I don’t have hardware GPU support for decoding the streams?
I had the same question and searched over hours. Seems to be more than changing one variable in a file. Changing the config file in /var/lib/docker/containers/FRIGATECONTAINER/config.v2.json does not work, its generated new every time the container starts. Guess we have to wait for a new version of the addon with the right env variable.
Maybe anyone can put a little light in the dark?
Thanks Mike
Maybe - my clips are from the 640x480 stream (same as motion detection). I’m on a pi4 4GB so I have been happy enough with those to not really try the high resolution. On my other pi running Magic Mirror it can’t do much with the 2k stream even at the lowest settings the 410 offers