Dev board and MCU for DVI in plus DVI out

[Psi]

What sort of dev board or MCU/FPGA would you guys recommend for a project which consists of basically...
- HDMI/DVI input  (doesn't need to work with HDCP, so no encryption issues)
- Frame by frame processing
      * Some visual functions, pincushion/rotation/fisheye correction etc...
      * Composite signal overlay, OSD etc...
- HDMI/DVI output (of the modified frames)
Resolution would need to be at least 720P

Is this even achievable with a BOM of less than $60

[mikeselectricstuff]

What sort of dev board or MCU/FPGA would you guys recommend for a project which consists of basically...
- HDMI/DVI input  (doesn't need to work with HDCP, so no encryption issues)
- Frame by frame processing
      * Some visual functions, pincushion/rotation/fisheye correction etc...
      * Composite signal overlay, OSD etc...
- HDMI/DVI output (of the modified frames)
Resolution would need to be at least 720P

Digilent's Atlys board Is probably the best fit.
NeTV may be another option, but doesn't have the fast DRAM so less likely to be useable for video effects.

Is this even achievable with a BOM of less than $60

Probably only in high (1K-10K) volumes.

[Psi]

Thanks

That Atlys board looks quite good, even has a demo project showing HDMI in/out with buffering of the data. On the other hand, the NeTV is cheaper and uses a cheaper FPGA

I might as well explain the project, as there might be a better solution someone can think of and i'd probably make it open hardware anyway.

What i'm thinking about is the oculus rift VR display (when it comes out) being used for FPV flying of RC planes.
You need a fisheye lens on the RC plane to see a wide angle and this image needs to be manipulated to match the 110degree field of view of the Oculus. Otherwise things will look very strange and looking around is likely to make you dizzy.

Initially the system would only need Composite input but eventually FPV will go HD and the camera signal is likely to be some form of DVI/HDMI. So planning for HDMI in/out is a good idea. Composite in/out is trivial to add.

[mikeselectricstuff]

Thanks

That Atlys board looks quite good, even has a demo project showing HDMI in/out with buffering of the data. On the other hand, the NeTV is cheaper and uses a cheaper FPGA

You'd probably need the external RAM to do any full-frame processing on video.

[Psi]

Yeah, at 720P/1080P+30FPS the throughput is crazy.

[krenzo]

Can the Raspberry Pi do it?  I've ordered one with plans to try driving the Oculus Rift using it, but I haven't received my board yet.

[mrflibble]

Digilent's Atlys board Is probably the best fit.

Seconded.

As for NeTV vs Atlys ... the NeTV only has an XC6SLX9 (Atlys has an SLX45 part), so you'd better do a rough estimate of your required resources.

[Psi]

Can the Raspberry Pi do it?  I've ordered one with plans to try driving the Oculus Rift using it, but I haven't received my board yet.

I have no doubts it could do HDMI out to the rift, its the HDMI input that's tricky.

[krenzo]

Oh right, it only has HDMI output and no input, and its GPIO pins certainly don't look to be differential or high speed capable to put an HDMI connector there.  I have the Atlys, but I haven't tried using any of its HDMI capabilities yet.

[Psi]

The Pi idea could work if i abandoned HDMI input.
It would be trivial to get a composite input working over usb in linux on the Pi.
Cheapish too

[Harvs]

I don't know about the PI.  Keep in mind it heavily relies on the GPU to do x264 decoding.  Last I checked there was no support for running user code on the GPU.

On the other hand, how about a uATX format PC (or laptop) with nvidia chipset or card and use CUDA to do the real time processing?  It would be great from a scalability perspective, if you run out of horsepower, just grab a bigger GPU off the shelf...

I have no real concept of just how hard the processing your talking about is to do in real time, but I'm assuming it'll take quite a lot of power.

Is this a one off project or something in quantity?

[krenzo]

I don't know about the PI.  Keep in mind it heavily relies on the GPU to do x264 decoding.  Last I checked there was no support for running user code on the GPU.

What do you mean?  I've seen people run Quake 3 on the Pi.  People have already created drivers/shaders to do view warping in support of the Rift, and one person did it with a purely geometric solution.  If you're implying you can't run pixel shader programs, then it could still be done by only rendering geometry.

[Harvs]

I don't know about the PI.  Keep in mind it heavily relies on the GPU to do x264 decoding.  Last I checked there was no support for running user code on the GPU.

What do you mean?  I've seen people run Quake 3 on the Pi.  People have already created drivers/shaders to do view warping in support of the Rift, and one person did it with a purely geometric solution.  If you're implying you can't run pixel shader programs, then it could still be done by only rendering geometry.

Of course, the output is the GPU.  Sorry my bad... 

I was just thinking to a few months ago I was looking at doing some GPU computing on the PI (massively parrallel programing like in the case of the CUDA solution), but that didn't seem possible.