Robust fast panorama stitching in mobile phones or cameras
View Patent ↗A camera that provides for a panorama mode of operation that employs internal software and internal acceleration hardware to stitch together two or more captured images to create a single panorama image with a wide format. Captured images are projected from rectilinear coordinates into cylindrical coordinates with the aid of image interpolation acceleration hardware. Matches are quickly determined between each pair of images with a block based search that employs motion estimation acceleration hardware. Transformation are found, utilizing regression and robust statistics techniques, to align the captured images with each other, which are applied to the images using the interpolation acceleration hardware. A determination is made for an optimal seam to stitch images together in the overlap region by finding a path which cuts through relatively non-noticeable regions so that the images can be stitched together into a single image with a wide panoramic effect.
1. A method for internally generating a panorama image on a mobile device, comprising:
capturing a plurality of images, wherein a first circuit is employed to accelerate a projection of each captured image into cylindrical coordinates;
employing a second circuit to accelerate determining a plurality of motion vectors for each overlap region between each captured image, wherein correspondences between consecutive images are based at least in part on the determined motion vectors;
based on said correspondences, determining pair-wise transformations to precisely align each pair of images and a global transformation to fit all images into a single rectangular panoramic image;
warping each image into the panorama image, based on said transformations; and
stitching each pair of consecutive warped images together along a determined stitch line between each corresponding pair of consecutive warped images to minimize visual artifacts, wherein the determined stitch Inc corresponds to a selected path of least energy that crosses an overlap region between the corresponding pair of consecutive warped images.
2. The method of claim 1 , further comprising enabling a display of the panorama image to a user.
3. The method of claim 1 , wherein the determined transformations are affine and are found by using linear regression.
4. The method of claim I, further comprising determining inliers and outliers among the correspondences between each pair of the plurality of captured images based at least in part on a Random Sample Consensus process.
5. The method of claim 1 , wherein the visual artifacts between each consecutive captured image are reduced at least in part by passing the stitch line through portions of the overlap regions of the consecutive images that are at least one of similar, without substantial edges, or without substantial detail.
6. The method of claim 1 , wherein the second circuit includes corner detection for improved reliability of a motion estimation process in the plurality of captured images.
7. The method of claim 1 , wherein the first circuit provides for interpolating a sparse grid of points with a look up table to accelerate the cylindrical coordinate projection for the plurality of captured images.
8. The method of claim 1 , further comprising estimating motion for the plurality of images based on at least a Mean Normalized Sum of Absolute Differences (MNSAD).
9. A mobile device for internally generating a panorama image of a scene, comprising:
an image capture device that is arranged to capture a plurality of images;
a first circuit that is arranged to accelerate a projection of each captured image into cylindrical coordinates;
a second circuit that is arranged to accelerate determining a plurality of motion vectors for each overlap region between each captured image, wherein correspondences between consecutive images are based at least in part on the determined motion vectors; and
a processor that is configured to enable actions, including:
based on said correspondences, determining pair-wise transformations to precisely align each pair of images and a global transformation to fit all images into a single rectangular panoramic image;
warping each image into the panorama image, based on said transformation; and
stitching each pair warped images together along a determined stitch line between each corresponding pair of consecutive warped images to minimize visual artifacts, wherein the determined stitch line corresponds to a selected path of least energy that crosses an overlap region between the corresponding pair of consecutive warped images.
10. The mobile device of claim 9 , further comprising a display for displaying the panorama image to a user.
11. The mobile device of claim 9 , wherein the determined transformations are affine and are found by using linear regression.
12. The mobile device of claim 9 , wherein the processor performs further actions comprising determining inliers and outliers among the correspondences between each pair of the plurality of captured images based at least in part on a Random Sample Consensus process.
13. The mobile device of claim 9 , wherein the visual artifacts between each consecutive captured image are reduced at least in part by passing the stitch line through portions of the overlap regions of the consecutive images that are at least one of similar, without substantial edges, or without substantial detail.
14. The mobile device of claim 9 , wherein the second circuit includes corner detection for improved reliability of a motion estimation process in the plurality of captured images.
15. The mobile device of claim 9 , wherein the first circuit provides for interpolating a sparse grid of points with a look up table to accelerate the cylindrical coordinate projection for the plurality of captured images.
16. The mobile device of claim 9 , wherein the processor enables further actions comprising estimating motion for the plurality of images based on at least a Mean Normalized Sum of Absolute Differences (MNSAD).
17. An article of manufacture including a processor-readable non-transitive storage medium having processor-executable code encoded therein, which when executed by one or more processor devices, enables actions for internally generating a panorama image of a scene on a mobile device, comprising:
capturing a plurality of images, wherein a first circuit is employed to accelerate a projection of each captured image into cylindrical coordinates;
employing a second circuit to accelerate determining a plurality of motion vectors for each overlap region between each captured image, wherein correspondences between consecutive images are based at least in part on the determined motion vectors;
based on said correspondences, determining pair-wise transformations to precisely align each pair of images and a global transformation to fit all images into a single rectangular panoramic image;
warping each imago into the panorama image, based on said transformation; and
stitching each pair of consecutive warped images together along a determined stitch line between each corresponding pair of consecutive warped images to minimize visual artifacts, wherein the determined stitch line corresponds to a selected path of least energy that crosses an overlap region between the corresponding pair of consecutive warped images.
18. The article of manufacture of claim 17 , further comprising enabling a display of the panorama image to a user.
19. The article of manufacture of claim 17 , wherein the determined transformations are affine and are found by using linear regression.
20. The article of manufacture of claim 17 , further comprising determining inliers and outliers among the correspondences between each pair of the plurality of captured images based at least in part on a Random Sample Consensus process.
21. The article of manufacture of claim 17 , wherein the visual artifacts between each consecutive captured image are reduced at least in part by passing the stitch line through portions of the overlap regions of the consecutive images that are at least one of similar, without substantial edges, or without substantial detail.
22. The article of manufacture of claim 17 , wherein the second circuit includes corner detection for improved reliability of a motion estimation process in the plurality of captured images.
23. The article of manufacture of claim 17 , wherein the first circuit provides for interpolating a sparse grid of points with a look up table to accelerate the cylindrical coordinate projection for the plurality of captured images.
24. The article of manufacture of claim 17 , further comprising estimating motion for the plurality of images based on at least a Mean Normalized Sum of Absolute Differences (MNSAD).