Home Forums 360° Stereo Panoramic Photography Stereo Consistency versus Seamless Stitch

  • This topic has 1 reply, 2 voices, and was last updated 5 years, 9 months ago by John Matulis.
Viewing 2 posts - 1 through 2 (of 2 total)
  • Author
  • #17914

    Thomas Sharpless
    Forum Member
    • Forum Posts: 33
    • ★★

    A typical problem with otherwise well-stitched stereo panoramas is variation of the general disparity/distance relation from place to place. You could also call it a variation of convergence angle. It is most evident in scenes that have large areas of close and distant subject matter. The cause is that the stitcher aligns close photos “more converged”, because close control points are “more diverged” than the distant ones. This reduces stereo disparity in close areas, compared to the correct disparity that would be seen if all photos were aligned at infinity. This effect is always evident on the ground near the camera. If you have achieved a nice seamless alignment of the pavement, it will be at the expense of removing most of the stereo disparity.

    Well, is there a solution? Yes, but not yet a good one. My PT3Dv1 helper for PTGui partially counteracts the loss of stereo in close areas, by adding synthetic control points that encourage PTGui to keep the original disparities present in the photos. However this conflicts with getting a perfect alignment in close areas, so there has to be a compromise. If you “wiggle” my stereo panos you will typically see some shear along seam lines in the foreground. I accept that misalignment as the price of more consistent disparity elsewhere.

    If you are serious about stereo panography, and have this problem, I urge you to try PT3Dv1. However a better solution is possible and I am working on it. This will involve aligning all the photos at infinity then pre-warping them according to local average depth. The final step, a warp-to-fit based on CPs near seam lines, will not have to distort the pre-warped photos so much. Most video array stitchers now use some similar process, and they get very good results. They use direct image matching (optical flow or stereo correspondence) whereas PT3Dv2 will depend on the depth information captured by control points linking stereo pairs.


    John Matulis
    Forum Member
    • Forum Posts: 24
    • ★★

    thomas, just wanted to say thanks for all your helpful info! much koodos

Viewing 2 posts - 1 through 2 (of 2 total)
  • You must be logged in to reply to this topic.