A Reconfigurable High-Res Network Camera

An anonymous reader writes "This technical article describes the architecture and design philosophy behind the Elphel Model 313, an intelligent 1.3 megapixel network camera that delivers full-sized images at up to 15 frames per second. The design of the Model 313, which has an embedded Linux computer based on an Axis ETRAX 100LX RISC processor, makes use of a reconfigurable Xilinx FPGA for much of the camera's internal control logic. Because both the embedded software and FPGA hardware algorithms are released as open source technologies, developers can readily customize the Model 313's operation to meet specialized requirements."

Comments

Slashvertisements work!

[The J Kid]

Yep. Its' offcial:

Slashvertisements work!
Just read this:

[..] and after it was mentioned on Slashdot my company (Elphel Inc.) was flooded with inquiries regarding general purpose network cameras

Re:Slashvertisements work!

[pjrc]

Yeah, it's a product. But....

It offers 1280x1024 at 15 frames per second. That's a LOT faster than other network cameras. That in itself is pretty newsworthy for slashdot.

It's also open source. The software, drivers, firmware and hardware HDL code are all open. Even if it didn't have incredible performance, this makes it pretty newsworthy for slashdot. The ability to actually tweak the hardware-level processing and compression of the camera data is intriguing.

The linked web page talks quite openly about the design process and how the thing really works (at least as a high level of abstraction). That in itself is pretty interesting and makes it fairly newsworthy for slashdot... at least as newsworthy as link to various writings describing how certain aspects of modern microprocessors work.

So, call it a "Slashvertisement", just because it's a product for sale and the author stands to sell some.

I think this is one of the coolest things slashdot has posted in quite some time. It's certainly a lot more interesting that yet another "sky is falling" story about privacy or copyright policy.

Great, can it be used with the HID program?

[AugstWest]

HID is the Human ID at a Distance program that DARPA is working on. Their goal is to develop technology to be able to positively ID individuals from a camera at a distance of 150 feet.

You can check it out here.

I wish digital cameras were more open...

[EnglishTim]

I've recently bought a Fuji S602Z - lovely camera, but there are a few little features it'd be nice to have that it doesn't have at present. I'm sure many of them wouldn't be too hard to code, but currently I just have to hope that Fuji will make the changes in any firmware update that they do.

An example is exposure gating. Currently my camera will take a series of three pictures, one normal, and two with slightly higher and lower exposure levels than the current setting. I'd like to be able to set this to five or more levels as it would be very useful for taking high dynamic range pictures. (You take a series of pictures at different exposures and combine them to produce a pictures that that a might higher range than a standard picture) Unfortunately, it's a bit of a niche need (useful largely only in post-production) and it seems unlikely that Fuji will implement it.

Re:FPGAs rule!

[Space cowboy]

Yes, they're cool.

They do not (and won't for some time) run anywhere near current CPU speeds. 200MHz in the sort of FPGA he's talking about takes very good knowledge of the internal FPGA architecture, and excellent HDL skills. There's a top end limit (not sure where it is, but it's about 3-400MHz in an FPGA that cost $2k)

There is no way you'll be reconfiguring your CPU any time soon, but having an FPGA resource on-hand would be useful. Reconfiguring for sorenson/MPEG/M-JPEG would be a neat trick.

FPGA's get their speed not from the clock as much as from their inherent parallelism - you can run each of the units slower, but have multiple units where it may not make sense in the general case.

Simon

Speed to burn at lower resolution frame rates

[CodeShark]

This is impressive work. As an avid follower of any technological improvement that can be used to lower the cost of creating high quality CAE (computer aided education), this looks like a winner -- but more so at a lower resolution than the top end of 1280x1024. Consider the author's text:

"There are a couple things that need to be cleaned up to fix that frame skipping, and then the camera will provide 15fps at 1280x1024 pixels, 60 fps at 640x480 pixels, and 240 fps at 320x240 pixels over the LAN connection."

Interpolating a bit, it seems reasonable that the camera could send 800x600 images at around 40 fps, which is a faster frame rate than standard NTSC video (60 half frames per second interpolated, 30 fps real), and at a resolution supportable by nearly all recent vintage consumer grade PC's, etc.

Granted, this is hardware speed, so encoding the massive data steam from the camera into a compressed but high quality playback format is another task for another machine or machines. But I'm still impressed.

As an example of why, take three of these, throw in some quality studio lighting, and come up with some editing software and hardware to mix the feeds together -- it looks to me like a person [with the technical knowledge to use the equipment and get good looking results] could create their own low cost production facility -- while still delivering image quality higher than is currently broadcast by most network and/or cable TV channels.