Mostly because of peoples here talks nearly allways about windows programming languages and 2D or 3D graphics which wont interest me :)

But cool place, love it.

E.K.Virtanen

www.ascii-world.com

I show the camera a pattern of points, then i define some points on the patterns as being at some specific location in 3D space, e.g. the first point on the pattern could be at location 0,0,0 the second at 1,0,0 etc. Objects are projected into the camera frame (image) with the equation:

(1)\begin{equation} x' = P * X \end{equation}

Where $x'$ is the point in the image, $X$ is a 3D location (these i define), and $P$ is the camera matrix (see also projection) which both includes the camera's 3D position and its orientation. I "just" need enough points to solve $P$ because i know $X$ and $x'$.

That sounds easy enough, but of course there is noise in the image, so what is on the image won't be perfect coordinates. The points are also quantized to the screen resolution which also adds noise, so in reality the equations are solved using singular value decomposition and some smart way of excluding outliers (points which have too much noise and only degrade the estimation of $P$).

This was just a little reading for those that are bored :) Btw, the reason we were stuck was not in estimating the position of the camera itself because this was actually just a function call in the OpenCV library, but we couldn't make sense of the coordinates it provided. Another problem we had was actually to find the 3D points in the camera image. Suffice to say it took a long time to find although it shouldn't have.

]]>:)

—hartnell

]]>Our problem was calculating the position of a video camera relative to the object it is viewing based only on what the video camera sees. 3D position and orientation. It's fun to work with when one makes progress :)

]]>Personally im working with source of yaBasic. Not with good succes tho. Also next issue of PCopy keeps me busy (beside of real life too)

E.K.Virtanen

www.ascii-world.com