Fisheye dewarp issues
AnsweredHi guys - I have a client with two fish eyes in a large warehouse. Really good and they cover the whole building really well.
The problem is with having live preview of the cameras. He wants to essentially dewarp a few locations on a layout - Its working BUT
A) Each dewarp of the same fisheye is downloading the entire video stream from the camera - Is there a way and if not as a developer suggestion shouldn't there be to have it download one fisheye video stream and render different dewarps off the one stream? I noticed he had a layout with 4 dewarps from the same camera and the network bandwidth dropped with each close of a view.
B) The CPU/GPU requirements to do this still seem to be very high. I was testing it on a CAD Workstation we just put together. i7 10700 with 64gb ram and a GTX1650 GPU. It didn't phase the pc much but it still used around ~16% cpu constantly and fluctuated constantly on GPU usage between 20% and 40% often near 30 but up and down.
I've also seen talk of a Use Hardware Video Decoding option but its not there??
-
Hi @...,
This is expected behavior.
When displaying a fisheye camera, it does use the full resolution or the primary stream and therefore there is also an additional load on the hardware and NICs. You could reduce this by increasing the compression but it might not resolve the issue entirely and it depends on the current settings.
A CPU load of ~16 - 40% doesn't worry me a lot and should be fine.
Regarding the Intel Quick Sync.
We introduced this as an experimental feature in our monthly patch of September 2020 (4.1.0.31767).
- KNOWN ISSUES
- Hardware decoding doesn't work on Ubuntu.
- Hardware decoding is disabled, if there are 2 graphic adapters and the primary is not Intel®. In this case users should set up a “primary display” to Intel® adapter to make hardware acceleration work.
Issues might occur, but our goal is to resolve these in or before version 4.3 if possible.
0 -
Thanks for the update - The load on that PC wasn't great but that's a hell of a spec for a PC to do nothing but run a TV in an office for monitoring showroom areas etc. I can recognise it takes some work to take a fisheye video and then flatten part of it and even more when you are rendering multiple flattened views from one fisheye. For the client its a big thing being able to cover one huge area with a single camera rather than having to put a camera in each location pointing multiple directions but the ability to render the live views is going to increase the costs massively to put pc's in powerful enough not to choke.
Perhaps downloading one full resolution stream from a fisheye but then rendering multiple lower resolution flattened outputs might lower the CPU utilisation as a development direction?? Different but similar I had a software TV recording application that was able to record an entire transport stream from a single frequency using one tuner and then demux separate recordings of the individual channels contained in the TS - It was similarly trans-formative in functionality.
I'm going to experiment a little with different hardware / gpu platforms and see if we can find a sweet spot where functionality works well without spending 2K per TV on the PC. I'll also see if Codec's make a difference - I think the Fisheyes are capable of H265 and H265 but I could be wrong.0 -
Hi @...,
Feel free to share your thoughts about improvements in the New Feature Ideas section.
This is the place where our product and design team will read along and engage.0
Please sign in to leave a comment.
Comments
3 comments