image corruption

[PeteJ]

This is from a Uniview camera on H.264 10fps, 1/5 key frame. I don't see this when looking at the web interface on the camera, but when I look at BI, this is pretty common at night, and happens from time to time during the day. Any idea what might be causing this? I am using decode acceleration on BI (Intel + VPP), not sure if that might be the reason?

 

[wittaj]

Yep turn off decode acceleration.

 

[PeteJ]

I noticed that the image (stills) quality seems to be less sharp with decode accel disabled, is that my imagination? Thanks

 

[wittaj]

Probably just imagination LOL or are looking at substream.

 

[PeteJ]

Disabling hardware accel worked. Doesn't seem to make the CPU work any harder either. The quality does seem poorer tho, i'm going to flip it back on and see if the quality improves.

 

[wittaj]

Quick sync/hardware acceleration isn't really needed anymore and it can be an energy hog.

Around the time AI was introduced in BI, many here had their system become unstable with hardware acceleration (hardware decode) (Quick Sync) on (even if not using DeepStack or CodeProject). Some have also been fine. I started to see errors when I was using hardware acceleration several updates into when AI was added.

This hits everyone at a different point. Some had their system go wonky immediately, some it was after a specific update, and some still don't have a problem, but the trend is showing running hardware acceleration will result in a problem at some point.

However, with substreams being introduced, the CPU% needed to offload video to a GPU (internal or external) is more than the CPU% savings seen by offloading to a GPU. Especially after about 12 cameras, the CPU goes up by using hardware acceleration. The wiki points this out as well.

Plus substreams opens up the possibility for older machines to be just fine, along with non-intel computers.

My CPU % went down by not using hardware acceleration.

Here is a recent thread where someone turned off hardware acceleration based on my post and their CPU dropped 10-15% and BI became stable.

But if you use HA, use plain intel and not the variants.

Some still don't have a problem, but eventually it may result in a problem.

Here is a sampling of recent threads that turning off HA fixed the issues they were having....

No hardware acceleration with subs?


Hardware decoding just increases GPU usage?


Can't enable HA on one camera + high Bitrate

And as always, YMMV.

 

[PeteJ]

Thanks @wittaj, I always learn a lot from your details replies.

I just tried the Intel hw decode (was using intel +vpp) and I was still getting the corruption, so I switched to Intel Beta just to see what happens.

Yeah, I was having some BI issues, and one of the first things support tells you to do is disable hw decode...

 

[IReallyLikePizza2]

I get this in Rekor Scout sometimes, with no decode acceleration

 

[wittaj]

I get this in Rekor Scout sometimes, with no decode acceleration

Rekor is outside of BI though.

 

[PeteJ]

Update: Intel Beta worked for about 6 hours before Miller Time. Back to CPU.

 

[Flintstone61]

What is the camera streaming data at? like 4096 Mbps or what? whats the substream data thruput?
this my old 2mp 3241ZAS . maybe you could up the substream so the console looks better?
What processor are you running and how many cams?

 

[PeteJ]

What is the camera streaming data at? like 4096 Mbps or what? whats the substream data thruput?
this my old 2mp 3241ZAS . maybe you could up the substream so the console looks better?
What processor are you running and how many cams?
View attachment 218702 View attachment 218701

I was at 2048, bumped it a few times to 4096 on a 1080p stream. It'll take me a day or so to see if it makes a difference.

I'm trying to improve stuff like this... (I believe this was when I had hw decode turned off)



This is my typical

 

[wittaj]

That looks like someone with a paper plate or the spray that makes it harder to read plates.

 

[PeteJ]

Could be. The sprays are usually not effective in daylight, but that could be a very dirty plate, too. I'll post the next one that I see that I know doesn't have a plate cover, paper, or dirt. It happens a few times a day.

 

[wittaj]

Yeah we aren't going to get every plate.

 

[IReallyLikePizza2]

Yeah we aren't going to get every plate.

 

[PeteJ]

Quick update. I've found that the default settings on the camera are terrible for LPR. The color saturation was too low for me to tell the color of the tags, and overall the image was too soft and probably contributed to the OCR error rate slightly. I've bumped color saturation, contract, sharpness up a lot and I think it helps. Bright color cars look super weird, but thats a small price to pay.

 

[IReallyLikePizza2]

Are you able to zoom in more? From the clips at the start it looks like you're quite far away

My plates appear much clearer than yours, because I'm almost all the way zoomed into the road

 

[PeteJ]

Are you able to zoom in more? From the clips at the start it looks like you're quite far away

My plates appear much clearer than yours, because I'm almost all the way zoomed into the road

View attachment 218850

View attachment 218849

I'm at 8.5x zoom, so I have a little left (10x) but at full zoom, there is a little more distortion and I am also using this to catch traffic in both directions, so more zoom means I am going to have to decide which traffic I want more...

 

[PeteJ]

ok... so too much goodness gives you some badness



Now OCR is reading the plate as Balifornia. Think about that for a sec... it thinks the cursive font that's 1/4 the size of the letters in the plate is the plate.

I wonder if Trump can rename my state to Balifornia. That would be pretty awesome.