User application video capture via kernel streaming?

Hello folks,

I've been hired for a in-house project (basicly an audio/video recorder,
which can also do some realtime effects and output the processed audio/video
in the while).
Since I'm an extremely "low level" guy, and it was up to me to choose the
hardware that would run my software, I implemented the audio part using the
kernel streaming method (studying DirectKS Microsoft sample, although I
rewrote everything from scratch).
This gives low latency, but I'd have followed this road even if it didn't
offer any such advantages over DirectSound. The reasons are simple, as I
said I'm a low-level coder (and I'm not specialized on Windows, rather,
often I work with proprietary hardware and OS), and I liked the more direct
interfacing with the hardware, which makes my programming life actually
easier and removes overhead (from my work and the CPU as well) and nonetheless
makes programming much more fun!

Now I've to write the video capture part, and I was wondering if an analogous
exists? What is the lowest-level possible when doing video capture from any
PCI card on Windows? It doesn't matter if, like the audio kernel streaming,
it won't be supported on Windows 2005.. because this box will always run
Win2K anyway, and on hardware chosen by me.

Some, even many of you may disagree with me, but I like the idea of doing
it the "lowest reasonable level possible", so I'd like to talk directly
with the driver (I don't pretend to talk directly with the hardware in
Windows, you see), and I'm already conscious of the disadvantages.

Please, could any kind soul direct me to the right path to follow? I'm
quite new to this kind of programming in Windows, and my battles with Google,
MSDN and DejaNews didn't shed much light on the subject.

Thanks!
Andrea