## finding normal vector to surfaces in video.

Feb 22, 2013 at 7:44pm

# finding normal vector to surfaces in video.

Hey all,
I hardly work with jitter but was interested in finding if there is a way to
approximate a normal vector of surfaces found in incoming video.

Now that i think about it, I am doubting this is possible.
LEt me know what you think

#66669
Feb 22, 2013 at 10:09pm

Hi leonard,

I have been exploring this concept for the past few days and I know that it is possible – just a question of methodology. Here’s some of my recent posts about this subject or somewhat related matters:

http://cycling74.com/forums/topic.php?id=45336

Let me know if you make any progress on this!

#239987
Feb 25, 2013 at 6:32am

ahh forsure I will.
I think it would be insanely difficult to accomplish this though…

lets say it used edge detection . then a wall lets say would look lie a trapezoid. the whatever direction it had to ‘shear’ the trapezoid to make a rectangle would be the direction or angle of the normal vector. ish something like that.
i am probably not spot on with that haha..
but well my question is how would it be possible to recognize a false vanishing point verses a true one.
consider a trapezoidal shape orthogonal to the camera. jitter ould think its actually a sheet slanted.
i think i would still be fun to figure something out like this though.

in a sense this is like making 3d from 2d video..

in algebra terms no transformation from 2d to 3d will be one to one. therefore it is obviously not possible but it could be interesting to play with some of this stuff.

I’m going to look into the links you posted…

appreciate the feedback . i hardly get any on cycling so i don’t spend much time on here.

thanks again

#239988
Apr 18, 2013 at 2:40am

something to do with Hough Transform

http://en.wikipedia.org/wiki/Hough_transform

#239989

You must be logged in to reply to this topic.