Auto align function & epiplar geometry & photogrammetry & computer vision

Wednesday, July 13, 2011

Auto align function is done (which is provided in PRO [paid] version).

Left and right images to generate a stereo image should be in epipolar geometry. Two images satisfied the epipolar geometry have same information (pixel values) in same row. It means there is no y-parallax. Existence of y-parallax makes viewer's eyes uncomfortable and that of x-parallax makes perception of height difference. The epipolar geometry is also one of fundamental concept for image matching since it reduce search space dramatically. The epipolar geometry is also used for generating maps since it provide height (3D information) from a pair of 2D images. If you want to know more, please read following three books.

   

First two books are from photogrammetry field which deals with aerial photogrammetry (all about mapping and precise positioning), satellite image processing (reconnaissance, etc.) also I am studying now. Third book is from Computer Vision field. I got the signature from one of the author (legendary Prof. Mikhail) of the first book (Modern photogrammetry) on the first page of my book.

There are many common things between photogrammetry and computer vision (Computer vision field is so wide; so when I say computer vision here, it means a small portion of computer vision field which deals with geometry of photography or imaging). In my opinion, photogrammetry have been more focused on quantity analysis such as precision of positioning in millimeter level while computer vision have been more focused on quality analysis such as what are in images. However, these days, differences between two field are getting smaller (but, still big). I am pretty sure that fusion of two field can produce much much more than what they can produce individually.

Anyway, the auto align function is not always perfect because image matching isn't always perfect. There are several techniques which reduce this imperfectness. In my opinion, computer always do things much faster than human; but, results are always less precise than what human did. So, please don't be mad if quality of resulting 3D pictures are bad; it happens almost always.

No comments:

Post a Comment