2

I have been given two images of a building.I know the transformation between these two images and I also know the coordinates of first camera position with respect to some coordinate frame , how can I get the camera coordinates of the second image in the same coordinate. I have been given nothing else information.

Dima
  • 38,860
  • 14
  • 75
  • 115
Mukesh
  • 117
  • 1
  • 1
  • 6
  • What do you mean when you say you know the transformation between the two images? Are you viewing a planar surface and you know the homography/projective function that maps one image to the other? Because generally to know the transformation between the images you need to know the geometry of the scene... – YXD Mar 27 '12 at 14:16
  • I have used step function of vision package in matlab to get the transformation between the two images.http://www.mathworks.in/help/toolbox/vision/ref/vision.geometrictransformestimator.step.html – Mukesh Mar 27 '12 at 14:41
  • Suppose I have an image of a building and I know the position of camera for this image with respect to some coordinate system. Now I took another image of the same building from another unknown location. Now from these two images, I have to find the camera position for the second image with respect to the same coordinate frame. – Mukesh Mar 27 '12 at 14:47

1 Answers1

2

If you have two images, then you can do the following steps to produce coordinates of second camera :

  1. Extract feature points(along with descriptor) of each image ( you can use SIFT features )
  2. Match feature points of both images( you can use flann library or brute force matching). Each of these matched pairs are tie points.
  3. You can use tie points to calculate relative position of one camera between another.( you can use 5-points method, but you need at least 5 tie points for this method-related paper).
  4. Use relative orientation(produced in step 3) with first camera's location for calculating second camera's location.

But you need some camera information like intrinsics for executing step 4 accurately, if you do not have such information there are methods for estimating them(like estimating the intrinsics of cameras)

Actually required implementation for each step can be found in very popular libraries like opencv and libmv(both are implemented in C,C++) but they may not have matlab wrappers for these method, you may find them if you like.

NOTE: By using this solution you can not use previously calculated image transformation result.

Community
  • 1
  • 1
AGP
  • 459
  • 1
  • 6
  • 20