Fast image processing using a beaglebone ??

[arbj2]

Hi,

Is it possible to quickly process images captured from a webcam connected to the beaglebone.

By processing I mean, the images have to be captured (via USB) or probably streamed, then the program has to scan the image (pixel by pixel) for a particular color say red.

When found an output is pulled high, and the entire process starts again.

For software I am planning to use OpenCV, and for experimental purpose I am using a canon S50 digital camera...later on I will get a high resolution web cam.

Please give your ideas....

thanks
a

I hope the above description does not sound vague...but as I do not have the specifics yet about rate of capture etc. I cannot give any values...I have used gphoto2

 

[andre_luis]

There are available interesting videos on the Web evidencing some successful experiments already performed on embedded platforms based on the same core processor ARM Cortex-A8 used on Beaglebone board.

As you asked, such projects employs OpenCV library to process captured video streams took from Webcam, but planned for achieve much more that color detection, as even online pattern tracking at rates close to many frames per seconds.

However, many of these demos was done using simple USB cameras with resolution in scale near to 1/4 VGA, what means that if you want to process higher resolution images, you should know that it implies on a processing increase on proportion of square, therefore decaying drastically the performance in matter of speed.

+++

 

[arbj2]

Yes I understand what you mean, the computer also needs a GPU for faster processing. But my application needs a rate of 5fps or even 2 fps.

I have also downloaded the opencv project demonstration for the beaglebone black from derek malloy's site. I have to study that...

thanks
a

 

[andre_luis]

Just to provide an additional information:

I´m not aware of Beaglebone, but at PCduino, even having a relative limited amount of resources I could even install IDE QTcreator on board and perform C++ code generation inside! It means that you will need not deal to complicated tasks involving cross compilation, but run directly.



+++

 

[arbj2]

cross compilation is no problem for me as I have made some projects using the bone.

What webcam do you use with the PCduino ???

thanks
a

 

[andre_luis]

I used the one referred here:

https://www.edaboard.com/threads/222694/


+++

 

[arbj2]

I have managed to do some image processing on my laptop using a Microsoft Lifecam VX-1000 webcam (very low resolution 640x480). I managed to develop a HTML5 and javascript program to read the LCD displays of digital multimeters and show the corresponding display as text.

I am now thinking how I can get the numeric data from the images more quickly..

thanks
a

 

[andre_luis]

Assuming invariance on scale and position of the target, the classic approach for such application is applying neural network pattern recognition algorithm.

Using the OpenCV library, the task to render images into raw files can be done easily but it´s not clear to me how could you integrate to Javascript, once it´s wrote in C langrage.

 

[arbj2]

I use a different approach. I identify the individual segments of each digit, so the program can determine which digit is being displayed.

The problem with this approach is that the segment, size, digit size, digit position etc. have to be setup first time. Once setup, the program works almost perfectly. But everytime the size of the image changes the system has to be setup again, apart from this drawback everything else is OK.

thanks
a

 

[andre_luis]

There are algorithms able to perform pattern recognition of targets placed in any scale and angle, but most of them are licenced, therefore you must read the terms of use. Even at OpenCV library you can find a nice example employing the SIFT algorithm.