Systems and methods for image capture
Images subject to panoramic image stitching are analyzed for suitability in three dimensional construction. Effects of translation changes between cameras are mitigated. The dioptric relationship of subject content of the images to the camera set identifies camera translation changes, and eligible camera position inputs, permissible to avoid parallax errors. Images within permissible ranges of each other are stitched to create panoramic images of subject content that a camera cannot fully capture with a single image. Registration plane selection based on translation distances between cameras, and permissible translation changes based on detected subject content depth are disclosed.
1 . A method for stitching a panoramic image, the method comprising:
receiving two or more candidate panoramic subimages;
for each candidate panoramic subimage of the two or more candidate panoramic subimages:
determining a camera position of a camera associated with the candidate panoramic subimage;
determining a distance between the camera position and points of the candidate panoramic subimage; and
classifying one or more points of the candidate panoramic subimage as far points based on the distance; and
aligning the far points across the candidate panoramic subimages by warping the candidate panoramic subimages to create a stitched panoramic image, wherein warping the candidate panoramic subimages comprises warping each candidate to a registration plane such that the far points from each of the candidate panoramic subimages align on the registration plane.
2 . The method of claim 1 , wherein classifying the points of the candidate panoramic subimage based on the distance comprises:
for each point of the points of the candidate panoramic subimage:
comparing the distance between the camera position and the point to a threshold distance value.
3 . The method of claim 2 , wherein classifying the points of the candidate panoramic subimage based on the distance comprises:
responsive to the distance between the camera position and the point being greater than or equal to the threshold distance value, classifying the point as a far point.
4 . The method of claim 2 , wherein classifying the points of the candidate panoramic subimage based on the distance comprises:
responsive to the distance between the camera position and the point being less than the threshold distance value, classifying the point as a near point.
5 . The method of claim 4 , wherein aligning the far points across the candidate panoramic subimages comprises:
disregarding near points across candidate panoramic subimages.
6 . The method of claim 1 , wherein warping each candidate panoramic subimage to the registration plane comprises:
warping the candidate panoramic subimage to another candidate panoramic subimage.
7 . The method of claim 1 , wherein aligning the far points across the candidate panoramic subimages comprises:
extracting one or more features associated with the far points;
matching extracted features associated with the far points across the candidate panoramic subimages; and
aligning corresponding matched extracted features across the candidate panoramic subimages.
8 . A system comprising:
one or more processors; and
one or more non-transitory computer-readable media storing instructions which, when executed by the one or more processors, cause performance of:
receiving two or more candidate panoramic subimages;
for each candidate panoramic subimage of the two or more candidate panoramic subimages:
determining a camera position of a camera associated with the candidate panoramic subimage;
determining a distance between the camera position and points of the candidate panoramic subimage; and
classifying one or more points of the candidate panoramic subimage as far points based on the distance; and
aligning the far points across the candidate panoramic subimages by warping the candidate panoramic subimages to create a stitched panoramic image, wherein warping the candidate panoramic subimages comprises warping each candidate to a registration plane such that the far points from each of the candidate panoramic subimages align on the registration plane.
9 . The system of claim 8 , wherein classifying the points of the candidate panoramic subimage based on the distance comprises:
for each point of the points of the candidate panoramic subimage:
comparing the distance between the camera position and the point to a threshold distance value.
10 . The system of claim 9 , wherein classifying the points of the candidate panoramic subimage based on the distance comprises:
responsive to the distance between the camera position and the point being greater than or equal to the threshold distance value, classifying the point as a far point.
11 . The system of claim 9 , wherein classifying the points of the candidate panoramic subimage based on the distance comprises:
responsive to the distance between the camera position and the point being less than the threshold distance value, classifying the point as a near point.
12 . The system of claim 11 , wherein aligning the far points across the candidate panoramic subimages comprises:
disregarding near points across candidate panoramic subimages.
13 . The system of claim 8 , wherein warping each candidate panoramic subimage to the registration plane comprises:
warping the candidate panoramic subimage to another candidate panoramic subimage.
14 . The system of claim 8 , wherein aligning the far points across the candidate panoramic subimages comprises:
extracting one or more features associated with the far points;
matching extracted features associated with the far points across the candidate panoramic subimages; and
aligning corresponding matched extracted features across the candidate panoramic subimages.