by JavaRN » Mon Aug 27, 2007 9:40 pm
by JavaRN
Mon Aug 27, 2007 9:40 pm
NovaOne wrote:
With your experience, can you think of any simple processing techniques i can do on board with a microcontroller?
Unfortunately I've never worked with microcontroller for image recognition I always worked with Personal Computers, infact with my Robonova all the processing (both image recognition and movement) is done on the laptop, the robot simply performs what the program in the laptop tells it to do. Any way if the microcontroller language supports arrays (in assembly it is a little bit more difficult) you can do whatever processing you like. Since microcontrollers have limited memory (sometimes enough to hold just one camera shot), you can down sample an image to 5x7 a technique used in OCR and then pass it through a Kohonen Neural net to get the corresponding character or shape. You can write simple programs on your PC to gather a database of down-sampled images, store them in an SD card and then use a card reader on your robot such as this one:
http://www.active-robots.com/products/accessories/ummc.shtml
to compare images using various techniques, including Neural Networks, BUT again, I prefer the processing to be done on the PC as resources such as RAM, Storage space, and processing power are less limited.
Oops, I was going to forget, there is also the gumstix which you can use, they are quite powerful and I'm sure that for you it isn't difficult to interface it with your Robonova. You can also connect the CMU cam to this small computer and you can perform all of the things I've mentioned. I haven't chosen this technique for two reasons -
1. it is expensive
2. I surely don't have enough space in my RN for a gumstix!!!
Charles
NovaOne wrote:
With your experience, can you think of any simple processing techniques i can do on board with a microcontroller?
Unfortunately I've never worked with microcontroller for image recognition I always worked with Personal Computers, infact with my Robonova all the processing (both image recognition and movement) is done on the laptop, the robot simply performs what the program in the laptop tells it to do. Any way if the microcontroller language supports arrays (in assembly it is a little bit more difficult) you can do whatever processing you like. Since microcontrollers have limited memory (sometimes enough to hold just one camera shot), you can down sample an image to 5x7 a technique used in OCR and then pass it through a Kohonen Neural net to get the corresponding character or shape. You can write simple programs on your PC to gather a database of down-sampled images, store them in an SD card and then use a card reader on your robot such as this one:
http://www.active-robots.com/products/accessories/ummc.shtml
to compare images using various techniques, including Neural Networks, BUT again, I prefer the processing to be done on the PC as resources such as RAM, Storage space, and processing power are less limited.
Oops, I was going to forget, there is also the gumstix which you can use, they are quite powerful and I'm sure that for you it isn't difficult to interface it with your Robonova. You can also connect the CMU cam to this small computer and you can perform all of the things I've mentioned. I haven't chosen this technique for two reasons -
1. it is expensive
2. I surely don't have enough space in my RN for a gumstix!!!
Charles
F'dan il-passatemp ghandek bzonn zewg affarijiet - FLUS u HIN. Zewg affarijiet li huma skarsi hafna u li jien minnhom ghandi vera ftit!