Em 7/10/2006, "James" <[EMAIL PROTECTED]> escreveu: >[EMAIL PROTECTED] wrote: > >>i am a mechanical engeneer student and i am beginning a project that >>consists in develop a tool for image processing for industry ( >>artificial vision). >> >> >True image processing algorithms are very different than what any von-neuman >architecture machine are designed for. That important fact aside, you >have to ask >yourself, are you going to be using the algorithms of others or >developing your own >(often phd Dissertation level of work here)? You should try to run some >canned >programs, form similar image processing probems on various machines >to guage your potential for success.
My idea is to develop one edge to another: from hardware to base software and image processing modules. > >>I remembered that i saw some nice arm processor boards that are not that >>difficult to build, and the price is quite low (about 1000 euros a >>prototype). >> >> >> >Arms are wonderful embedded processors, but they might not be the best >architechture for the particular family of image process algorithms you >are going to run. Furthermore, you can always grab images with one >board/system and process the data (run the image processing algo) on >a DSP, multiprocessor system or a GPU? the images are to recognize objects, for example, letters, in an image. This is a very easy example. They are not for video, or to take nice pictures :P > >>My question is: is the arm processor capable of supporting multiple >>acquiring image devices (cameras) and process them (by a C program)? >> >> >> >Dunno, find one on the net and test drive it. People usually select the >embedded >processor with the added hardware features that make their solution to a >problem (robust). > >>I think an intel pentium @ 450 MHz is capable, but how do i comparfe with >>arm, before i buy? >> >> > >Run an image processing algo on both and collect/analyze the results, >runtimes, >memory usage etc..... google for benchmark ..... > >>My idea is to develop an custom board and to connect there multiple >>cameras (as needed ... 1 to 4 cameras) filter the image and provide >>results (for example, to count pills in a pill box), basing that in >>installing a base debian system and then provide packages of the >>application of image treatment. >> >> > >A quicker, cheaper, and cooler idea would be to port an existing image >processing >algo to a GPU, or develop an image processing algo from scratch on a GPU. > that's the thing i want, but ... GPU? why not something like having a linux system and develop applications on top of it? I have this idea, because target systems have windows CE (or embebed) installed ... and i have much more features in linux, for developping at this level. >After all GPUs are designed from the ground floor up to process image, >graphics >and video in specialized hardware inherent to the GPU. (SIMD, MIMD) >multiport >ram. For example the ram(memory) in a high end GPU is the fastest memory >currently availalbe. It smokes the standard memory or chache performance >of risc/cisc processor. > >Check this out: > >http://developer.nvidia.com/object/cg_toolkit.html >http://graphics.cs.uiuc.edu/~jch/ >http://gpu.sourceforge.net/thesis-ideas.php >www.*c*s.sunysb.edu/~vislab/papers/GPU*c*luster_SC2004.pdf >www.doc.ic.ac.uk/~phjk/Publications/POHLL06-AcceleratingWithGPU.pdf > i'll look at them. Like i said ... i am a mechanical engeneer, just looking for some base on ideas. As i am reading by your comments, arm arquitecture is just too slow ... and i wish to thank you all for your ideas. >googling for: gpu C +image +processing > >yeilds many fabulous results..... > >HTH, > >James > > > >-- >To UNSUBSCRIBE, email to [EMAIL PROTECTED] >with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED] >
