How do you build a usable DVR system?

[Nimrod 0]

I have been researching the use of DVR cards in a computer to use for surveillance. And after reading this document here.

 

http://www.initsys.net/attachments/Compression%20and%20DigitisationPDF.pdf

JPEG2000, MJPEG, MPEG and H.264 in the security environment.

By Christopher Berry.

 

It seems it is next to impossible to put a useable system together. With all the manufactures jumping on the H.264 bandwagon there is no Jpeg, Jpeg 2000, or wavelet DVR boards left that I could find. Geovision dumped their wavelet at the end of 2006

 

Essentially what the article points out is that H.264 is a video standard for displaying motion. And in 1 second of video there are only 2 frames with the full video information. This means if you record at 30 FPS you can only play back 2 of the frames in one second of recording going frame by frame for identification, and this is far to infrequent to catch much happening even at a slow walk. In my opinion 7 or 8 FPS is barely usable. Also Jpeg is required for legal court video, another reason H.264 is bad.

 

I was hoping to find a hardware accelerated card with some form of Jpeg support that would de-interlace the NTSC image and give me 640x480 or D1 output. With my choice of 10 to 30 FPS and up to 16 cameras.

 

In the last week I have read a lot of posts on this board. There are clearly a lot of informed people here. How are you getting around this problem?

 

 

Thank you

Nimrod

[Soundy 1]

I think the problem is overstated. It's true, you only get a complete picture for every I-frame, and a P- or B-frame won't render a full picture on its own. However, if a DVR is properly decoding the stream, it should only need to see the relevant I-frame in order to fully reconstruct the P/B-frames.

 

Think of it this way: you have a video of a city street with someone walking through it. The system snaps an I-frame of the entire scene. Then for the next frame, it looks at the picture, determines that only the person and the area immediately around him has changed, and so stores that information and dismisses everything that hasn't changed. Repeat this several times, then take another complete snapshot for reference.

 

As long as the DVR can read at least one I-frame, the entire image can be reconstructed, and advanced frame-by-frame, by displaying the complete picture, and then drawing in the changes.

 

The only real danger lies in an I-frame being dropped, but the likelihood of JUST the I-frames being lost is slim and no greater risk than any other frames being lost.

 

Long story short: don't get too worked up about it My experience with H.264 playback at lower framerates is, it does frame-by-frame just fine.

 

The author is accurate on one point: 30fps is really over-rated. Most people couldn't even tell the difference between 15 and 30, and most wouldn't realize there was anything particularly wrong with even 10fps. See http://www.panasonic.com/business/security/demos/PSS-recording-rates.html

[Nimrod 0]

“Long story short: don't get too worked up about it My experience with H.264 playback at lower framerates is, it does frame-by-frame just fine.”

 

OK that clears up some misconceptions I had. From the article I thought all frames were thrown out except for the 2 key frames for frame by frame playback. I also found some white papers from several universities on Jpeg 2000 vs H.264 and there is no clear winner. The best image depends on various circumstances not what codec was used. I suspect what is going on here is that H.264 is “good enough” and has far superior compression than Jpeg 2000. This solves the big problem of storage size for larger systems and newly emerging megapixel cameras.

 

Nimrod

[Soundy 1]

I also found some white papers from several universities on Jpeg 2000 vs H.264 and there is no clear winner. The best image depends on various circumstances not what codec was used. I suspect what is going on here is that H.264 is “good enough” and has far superior compression than Jpeg 2000. This solves the big problem of storage size for larger systems and newly emerging megapixel cameras

Hmmm, yes and no. This topic has actually been discussed here extensively.

 

The thing is, H.264 really benefits at 30fps, but as framerate goes down, so does the benefit over other codecs as far as storage space and bandwidth.

 

They're also comparing to "standard" codecs, but many manufacturers have their own proprietary recompression systems that can equal or even better H.264 under many conditions, with less loss of quality. Check out this article, for example: http://www.3xlogic.com/aztech

 

And yes, you're correct, different types of scenes and circumstances see different results with different codecs; there isn't one PERFECT codec for every instance.