Video Anayltics and compressed video.

[johrt]

Can video analtics be performed on compressed video. What I have read seems to indicate that it can not. What would be some of the ptifalls if analytics were able to be performed on compressed data. If compressed data is OK, what format is preferred?

[Steve]

Are you talking about prerecorded video?

[johrt]

Could be. But what I am concerned with is normal digital video using jpeg4 or other simlar compression. The thought is that I would perform any required analytics from a centralized location.

[Steve]

Well I guess it would depend on where you are recording the video from. An IP Camera or Video Server would be taking the video and compressing it before the video analytics kick in. So my answer would be yes, video analytics can be performed on compress video.

I'm not 100% certain though as I don't fully understand the question.

What would you be recording the video from? IP Camera, CCTV Camera or Video Server?

[chad_cooper]

To my knowledge, video analytics must be perform prior to compression or the video must be decompressed, then the analytics can run, then recompressed and saved. Different manufacturers do some very neat tricks to help save CPU load during this time.

If this is at the camera, it is much easier to do the analytics prior to compression.

I think you where talking about MPEG4 compression.

Hope it helps

[fwboyd]

Depends on what you mean compressed.

All digital video has some loss when converted from analog, so the particular codec is key - along with resolution.

We have been writing video analytics for our traffic market using H.264 video, which is "compressed" about 3X over MPEG2 for example.

Our experience suggests that camera quality (TVL count), resolution (we use full D1) , and lens selection (wide angle warps objects) can affect success rate, but H.264 is not a contributor.

[GregFisk]

I am in agreement with FWBoyd, and Chad Cooper for the most part. Although I would say that the issue is only about pixels on target. Each manufacturer of algorithms requires some number of pixels on target to achieve the desired result (they will tell what that is if you ask them). If they do not get these pixels they cannot perform always more is better. In the case of MPEG4 (not JPEG4 by the way)and H.264 (which is MPEG4 version 10) it is a differential compression structure consisting of I frames (complete frames) and P frames differential information (information that has changed from the previous frame), In addition to this inter-frame compression there is also intra-frame compression often referred to as “quality”. Chad if you are saying that the frame must be reconstructed to a single frame in order to amass the pixels required this may or may not be true. When you think about it video analytics are based on advanced motion analysis so for a rule to be broken (in most cases) some sort of motion must have occurred. Where the motion has occurred will create new pixels since we are only interested in this area for analysis these are the only pixels that matter. Therefore reconstruction or decoding of the stream is not always necessary. On the other hand if the quality or compression as it called in Axis cameras is set too high it will cause molting at sharp transitions and will make life much more difficult for the analytic. Once this has been set though, there is no recovering it.
In regard to the source of the stream, IP camera or encoder Steve this is not accurate at all. Think about it an IP camera is nothing more than an analog camera with an onboard encoder. Yes the digitized video comes to the encoder digitally in the camera right out of the DSP so there are some advantages to this method and they may apply to analytics but it has to do with DTVL and that is something the camera/encoder manufacturers do not want to publish. This is a little comlicated and if you wish I will explain in more detail ina later post just ask.
There are only 4 components to a video recording solution, it does not matter if it is analog, hybrid or IP they all consist of the same parts. They are an accumulation device (camera), encoder (may be in the camera, may be stand alone, or it may be in the DVR) Video management software, and storage.
Analitics should be able to be accomplished using MPEG2 MPEG4 and H,264 in it's native format.