0

Given two images of same object from two views.How to decide which one is left view and which one is taken from the right side. I have calculated the corresponding matching points and have removed the outliers by using fundamental matrix calculated with the help of Ransac. Using these corresponding matching points in two images I am deciding which one is taken from right side and which one from Left side by simply comparing their x cordinated in image.I am not sure whether my approach is correct in general. Can anyone suggest any other way to determine the position of the camera(i.e. left or right.)

Optimus 1072
  • 309
  • 4
  • 18
  • Generally, left view has larger x. – Yang Kui Jan 26 '16 at 07:12
  • Yeah, This approach is correct for most cases but for not all.Consider the cases where in left view, the object under consideration is such that it lies on the left side of the image then the algorithm is not working properly. – Optimus 1072 Jan 26 '16 at 07:17
  • Can you post the images where the "larger x" rule failed here? – Yang Kui Jan 26 '16 at 07:34

1 Answers1

1

If your camera is calibrated, then you can compute the Essential matrix from the fundamental matrix, and then decompose it into a rotation and translation between the two views, as shown in this example. This will tell you whether the second view is on the left or on the right.

Dima
  • 38,860
  • 14
  • 75
  • 115
  • Thanks for your answer. I also considered this but going for 3-D reconstruction(since there are four possible camera orientation and out of them only one is correct ) doesn't seem a feasible approach to me, since we just need to tell which one is right and which one is left view .I am looking for a solution where we dont need to traingulate. – Optimus 1072 Jan 25 '16 at 14:54
  • It is actually not so bad. If performance is a concern to you, you can use the mid-point triangulation algorithm. It is not as precise, but it is quite fast. And all you really need to know is whether the points end up behind one or both cameras. If you are using MATLAB, `cameraPose` will do all of it for you. – Dima Jan 25 '16 at 15:15