#Morning folks.... QQ... I am getting car
severe sequoia
Set a max_area filter for car
spare flame
Is that retroactive like annotation offset?
severe sequoia
it will just filter out what gets detected as a car in the future
essentially ignoring any car detections that are larger than that size
so if you specify a size that no car would be, it will stop the detection that you are seeing
what model / detector are you running?
I can also make some other recommendations about the config, some things are not optimal
spare flame
what model / detector are you running?
using coral.ai for detection, and yes... I would love feedback / critique on my config if you're offering https://gist.github.com/johntdyer/811f22e0144b137d23c4ff38ef6aad7d
severe sequoia
I'll just make a list, let me know if anything is unclear
- You have the nvidia hwaccel preset set but then are adding manual hwaccel args to input args, this is wrong and going to increase cpu usage. I'd suggest just using one of the input_args presets
- no reason to override global_args, you are already missing out on changes that decrease cpu usage
- The driveway camera is presumably using a large mask to block traffic which is discouraged in the docs and I'd suggest just using required zones for events. Frigate is designed to efficiently track objects across the frame
- input args at the driveway camera level again are removing important args and duplicating hwaccel
- Just a heads up that experimental ui blocks the new live ui types
spare flame
@severe sequoia this looking better? https://gist.github.com/johntdyer/c7b88b312b82c1070e44f7803bfd422c still considering your suggestiojn on #3
severe sequoia
anothing thing I'm noticing is a secondary line for a log of cameras with #video=h264
aren't those cameras already h.264?
those are likely not needed
you have
ffmpeg:
output_args:
record: preset-record-ubiquiti
set for every camera but it is set globally so that is not needed
I would also recommend setting input_args: preset-rtsp-restream globally
spare flame
So I have mostly unifi cameras but moving to dauha cameras ( currently 1 ptz )
I meant to set it for each camera and set it to the non-unifi preset globally … guess I got it reversed 🙂
spare flame
aren't those cameras already h.264?
Forgot how I ended it with this config item, I’ll retest and see if still required for my deployment
spare flame
hey @severe sequoia , so I switched to zones as you suggested for the driveway camera and now I finally understand what was meant by "lower middle of bounding box " and the inertia setting... so there is this space between the driveway and the street that due to the angle of the camera a bounding box is almost always going to be in driveway zone... Would the best route to combat this be to to play w/ the interia settings, move the zone up from the street and sacrifice some of the driveway zone, or use a mask or something ??
severe sequoia
image would be helpful to see
spare flame
yea, working on that part 🙂 always get the new line key wrong on discord when drafting a message
three zones, I think they are self explanatory but let me know if not
basically the inverse of the motion mask on the street you mentioned might be problematic.. i zoned out the camera into three zones and did detection only in those, excluding the street
hope that makes sense , and thank you very much for your help
severe sequoia
yeah that zone is basically on the street, I would recommend bringing it lower on that top part
spare flame
k, I'll give it a shot
severe sequoia
any improvement?