PDA

View Full Version : Discussion hardware for vision processing?


meng.wei
Sep 23, 2008, 08:50 PM
I am working on a UAV project that requires the identification of fires, forest fires from the air to be precise. While I'm not very familiar with the software aspect of things, I believe I can learn it easily. What I am having trouble with right now is selecting a hardware platform to run it on.

There are two video feeds, both at 640x480 through RCA connector.

Initially I thought I could use FPGAs, they are supposed to be very good at image processing, and I have some limited HDL experience, but it turns out its extremely difficult to find any online resources on how to write image processing code in HDL. and since I don't have a masters degree in computer science I don't think I can learn it in time. Unless of course someone here has done it before and could offer me with some hints?

Then there is the good old uC/processor, and this is where I'm stuck: what speed am I likely to need? ATMega128 with 16MIPs might be enough, or it might not. Do I need something more powerful like an ARM7 processor? or could it be done with something smaller? Frame rate of around 5FPS should be sufficient.

How much memory am I going to need? both internal and external, if any?

And lastly, what kind of I/O ports or hardware are required to convert an RCA feed into something the processor could handle?

Great thanks to anyone with helpful hints and suggestions!

Tomapowa
Sep 24, 2008, 10:36 AM
Probably not exactly what you are looking for but some good reading...

http://geo.arc.nasa.gov/sge/UAVFiRE/completeddemos.html

If you are an IEEE member, there's a ton of info on this exact science...
as a few examples:

http://ieeexplore.ieee.org/iel5/9716/30678/01421401.pdf?arnumber=1421401

http://csdl2.computer.org/persagen/DLAbsToc.jsp?resourcePath=/dl/proceedings/&toc=comp/proceedings/alpit/2008/3273/00/3273toc.xml&DOI=10.1109/ALPIT.2008.49

http://csdl2.computer.org/persagen/DLAbsToc.jsp?resourcePath=/dl/proceedings/&toc=comp/proceedings/cvpr/2007/1179/00/1179toc.xml

Typically, you would convert captured frames to MPEG (real-time) and process it from there...

Here's also something to get you thinking...
http://cnx.org/content/m11987/latest/
http://cnx.org/content/m11988/latest/
http://cnx.org/content/m11989/latest/