Stereo vision: How to do it?
10 years
Hello again everyone.

I'd like to ask a few questions and some advice on Stereo vision. Mathematically, what happens inside a computer when its processing stereo vision. What data does a computer manipulate and put together to utilize stereo vision? How do I make this happen in my robot using roborealm, as I am considering it for my project, NINA.

Thanks everyone!
Anonymous 10 years
For a very crude description of what stereo correlation is: basically the computer will compare each image pixel by pixel to the other image and determine given a pixel and its immediate surrounding pixels how much that pixel has moved with respect to the other image. The surrounding pixels are used to provide context as just using a single pixel value for comparison is normally not sufficient. Note that in stereo correlation this comparison normally happens along a horizontal line as one assumes the images are simply shifted in the X direction.

The pixel to pixel comparison is what makes stereo vision so slow. Comparing each pixel and its neighborhood between two images can require millions of comparisons.

Currently there is no way to create a stereo map in RoboRealm ... unless you use the movement detector module to create a very VERY crude approximation of depth disparity. We are currently researching available techniques for stereo vision and while we have made progress we do not have anything sufficient to release at this time. i.e. the current technique is too slow and imprecise enough to be used for anything useful.

mbahr from United States  [23 posts] 7 years
Any progress on stereo vision or structured light modules?


This forum thread has been closed due to inactivity (more than 4 months) or number of replies (more than 50 messages). Please start a New Post and enter a new forum thread with the appropriate title.

 New Post   Forum Index