Ask Your Question

Revision history [back]

click to hide/show revision 1
initial version

Improving OpenCV performance when stitching 360° mosaics

Hi,

I am trying to compose a 360° mosaic using OpenCV. However, I am hitting a major performance issue due to the following.

During the warping process of all my images, due to the 360° nature of the stitching, some images are bound to be "cut" in two parts, like this:

Wrapping Image

As a result, OpenCV registers this image as being a 1806x365 image after warping, which means that during most following steps (especially seam finding), the performance is drastically decreased. Indeed, when finding overlapping images, this particular image is overlapping with every other image since it is now sizing the entire canvas size.

Ideally I would love to find a way to stitch as if the picture were not wrapped on both sides of the screen:

Non wrapping image

Anyone has any experience doing this?

Thanks!

Improving OpenCV performance when stitching 360° mosaics

Hi,

I am trying to compose a 360° mosaic using OpenCV. However, I am hitting a major performance issue due to the following.

During the warping process of all my images, due to the 360° nature of the stitching, some images are bound to be "cut" in two parts, like this:

Wrapping Image

As a result, OpenCV registers this image as being a 1806x365 image after warping, which means that during most following steps (especially seam finding), the performance is drastically decreased. Indeed, when finding overlapping images, this particular image is overlapping with every other image since it is now sizing the entire canvas size.

Ideally I would love to find a way to stitch as if the picture were not wrapped on both sides of the screen:

Non wrapping image

Anyone has any experience doing this?

Thanks!

Edit: I realized it's not clear why this image has to be cut in half. The picture is part of a full 360 panorama, as it can be shown here

Full Panorama

Improving OpenCV performance when stitching 360° mosaics

Hi,

I am trying to compose a 360° mosaic using OpenCV. However, I am hitting a major performance issue due to the following.

During the warping process of all my images, due to the 360° nature of the stitching, some images are bound to be "cut" in two parts, like this:

Wrapping Image

As a result, OpenCV registers this image as being a 1806x365 image after warping, which means that during most following steps (especially seam finding), the performance is drastically decreased. Indeed, when finding overlapping images, this particular image is overlapping with every other image since it is now sizing the entire canvas size.

Ideally I would love to find a way to stitch as if the picture were not wrapped on both sides of the screen:

Non wrapping image

Anyone has any experience doing this?

Thanks!

Edit: I realized it's not clear why this image has to be cut in half. The picture is part of a full 360 panorama, as it can be shown here

Full Panorama

If I remove the 16th image, it takes only 4 seconds to stitch the full panorama. Adding that 16th image adds a whole 20 seconds to the process, simply because the algorithm is trying to find overlaps between this 16th image and all 15 others.