Transforming Images to a Projective Plane: What Algorithm Should I Use?

[todorangrg]

Hello,
I'm working right now on a project which evolves transforming an image taken with a camera to a plane surface at an angle x and a height h into a view-from-the-top image. I searched which algorithm should I use and I saw that it is a matter of projective plane transformation but there I got stuck because i don't now how to express the "viewer point" if I want the real image to be constructed plane, as a map. Or should I search for something else? What should I use?

your help would be highly appreciated.
George T.

 

[futurebird]

Can you post a diagram. I'm having a hard time understating what you're asking?

For example: "at an angle x" were is this angle?

I think you should draw the set up you have with the camera, and label the "height h" and "the angle x" so we know what you mean.

 

[todorangrg]

I mean that the camera is standing on a car that is on a flat road and the perpendicular line from the road to the camera has the height h. For a better viewing the camera pithced in a little downloard position(in order tu be able to se the things that are close to it and not just the infinite points) the view of this camera(perspective) I want to transform in a map like view.
I hope I made my self clear...if not just tell me and I will post a diagram.

 

[futurebird]

Okay, so you're looking at taking a moving image and making it into a map? Or are you using photos?

In either case you'll need to do a lot of photo stitching. There are some 3rd party softwars that do similar tasks, using still photos:

http://iovision.com/demyst/qtvr_panorama.html

http://www.apple.com/quicktime/technologies/qtvr/

http://www.ptgui.com/info/image_stitching.html

Now how your image looks will depend on your lenses. To some degree many have a "fish eye" effect-- this must be adjusted to get the images to stitch together seamlessly.

In other words it's not just a matter of simple geometry. But happily there are programs out there that address these very problems.

 

[todorangrg]

In fact I have a tracking electronic camera and I want to monitor the black road(the road will be "seen" by the camera as 8 pixel height blobs of which coordinates I know in a 144*80 pixels camera so I don't think that I need a very precise corection because the rezolution of the cam itself will be low). After I make the map I create with this blobs B spline curves using the blobs as contol points...but that's something else. Anyway the stiching of the pictures doesn't bothers me because I can monitore the movement of the car in order to know how much the picture(in fact the road line because that's the only thing that matters to me) advanced.
So...what matematical function should I use so that each dot from the image to have a corespointing dot in the "map"?