Three-dimensional dense structure from motion with stereo vision
Disclosed examples include three-dimensional imaging systems and methods to reconstruct a three-dimensional scene from first and second image data sets obtained from a single camera at first and second times, including computing feature point correspondences between the image data sets, computing an essential matrix that characterizes relative positions of the camera at the first and second times, computing pairs of first and second projective transforms that individually correspond to regions of interest that exclude an epipole of the captured scene, as well as computing first and second rectified image data sets in which the feature point correspondences are aligned on a spatial axis by respectively applying the corresponding first and second projective transforms to corresponding portions of the first and second image data sets, and computing disparity values of a stereo disparity map according to the rectified image data sets to construct.
1. A method comprising:
determining relative camera positions of a single camera associated with first image data set obtained at a first time and a second image data set obtained at a second time;
computing a first projective transform and a second projective transform based in part on the determined relative camera positions, the first projective transform corresponding to a first region of interest of the first and second image data sets and the second projective transform corresponding to a second region of interest of the first and second image data sets, wherein both the first and second region of interests of the first and second image data sets excludes an epipole;
computing first rectified image data set based in part on the first projective transform;
computing second rectified image data set based in part on the second projective transform, wherein corresponding points in the first rectified image data set and second rectified image data set are aligned on a spatial axis;
calculating disparity values for a pixel location using the computed first and second rectified image data sets; and
generating a disparity map image using the calculated disparity values for the pixel location.
2. The method of claim 1 , further comprising computing depth values for the pixel location of the scene according to the disparity values.
3. The method of claim 1 , wherein determining the relative camera positions includes:
computing the feature point correspondences between the first and second image data sets, the feature point correspondences individually correlating a pixel location of the first image data set and a pixel location of the second image data set for which the image data match; and
computing an essential matrix that characterizes relative positions of the single camera at the first and second times according to the feature point correspondences.
4. The method of claim 1 , wherein the regions of interest include a right region, a lower region, a left region, an upper region, and a central region that includes the epipole, the central region disposed vertically between the lower region and the upper region, and the central region disposed laterally between the right region and the left region.
5. The method of claim 4 , further comprising computing the pairs of first and second projective transforms for only the right region, the lower region, the left region, and the upper region.
6. The method of claim 5 , further comprising computing the first and second rectified image data sets for only the right region, the lower region, the left region, and the upper region.
7. The method of claim 4 , further comprising computing the first and second rectified image data sets for only the right region, the lower region, the left region, and the upper region.
8. The method of claim 4 , wherein the first and second projective transforms are computed to compensate rotation of the single camera between the first and second times to facilitate virtual rotation of the image data from the single camera in computing the first and second rectified image data sets.
9. The method of claim 1 , wherein the first and second projective transforms are computed to compensate rotation of the single camera between the first and second times to facilitate virtual rotation of the image data from the single camera in computing the first and second rectified image data sets.
10. The method of claim 1 , further comprising determining the relative camera positions using a processor, and computing the projective transforms and the rectified image data sets using a hardware accelerator circuit.
11. An imaging system, comprising:
a single camera to obtain a first image data set at a first time and a second image data set at a second time; and
a video processor configured to:
determine relative camera positions of the single camera associated with the first image data set and the second image data set,
compute a first projective transform and a second projective transformed based in part on the determined relative camera positions, the first projective transform corresponding to a first region of interest of the first and second image data sets and the second projective transform corresponding to a second region of interest of the first and second image data sets, wherein both the first and second regions of interest of the first and second image data sets excludes an epipole,
compute first rectified image data set based in part on the first projective transform;
compute second rectified image data set based in part on the second projective transform, wherein corresponding points in the first rectified image data set and second rectified image data set are aligned on a spatial axis,
calculating disparity values for a pixel location using the compound first and second rectified image data sets; and
generating a disparity map image using the calculated disparity values for the pixel location.
12. The imaging system of claim 11 ,
wherein the processor is programmed to determine the relative camera positions; and
wherein the video processor further includes at least one hardware accelerator circuit to compute the projective transforms and the rectified image data sets.
13. The imaging system of claim 11 , wherein the video processor is configured to determine the relative camera positions by:
computing a the feature point correspondences between the first and second image data sets, the feature point correspondences individually correlating a pixel location of the first image data set and a pixel location of the second image data set for which the image data match; and
computing an essential matrix that characterizes relative positions of the single camera at the first and second times according to the feature point correspondences.
14. The imaging system of claim 11 , wherein the regions of interest include a right region, a lower region, a left region, an upper region, and a central region that includes the epipole, the central region disposed vertically between the lower region and the upper region, and the central region disposed laterally between the right region and the left region.
15. The imaging system of claim 14 , wherein the video processor is configured to compute the pairs of first and second projective transforms for only the right region, the lower region, the left region, and the upper region.
16. The imaging system of claim 14 , wherein the video processor is configured to compute the first and second rectified image data sets for only the right region, the lower region, the left region, and the upper region.
17. The imaging system of claim 14 , wherein the video processor is configured to compute the first and second projective transforms to compensate rotation of the single camera between the first and second times to facilitate virtual rotation of the image data from the single camera in computing the first and second rectified image data sets.
18. A non-transitory computer readable medium, comprising computer executable instructions which, when executed by a processor, cause the processor to:
determine relative camera positions of a single camera associated with first image data set obtained at a first time and a second image data set obtained at a second time;
compute a first projective transform and a second projected transform based in part on the determined relative camera positions, of the first projective transform corresponding to a first region of interest of the first and second image data sets and the second projective transform corresponding to a second region of interest of the first and second image data set, both the first and second regions of interest of the first and second image data sets excludes an epipole;
compute first rectified image data set based in part on the second projective transform;
compute second image data set based in part on second projective transform, wherein corresponding points in the first rectified image data set and second image data set;
calculating disparity values for a pixel location using the computed first and second rectified image data sets;
generating a disparity map image using the calculated disparity values for the pixel location.
19. The non-transitory computer readable medium of claim 18 , further comprising computer executable instructions to cause the processor to:
compute the feature point correspondences between the first and second image data sets, the feature point correspondences individually correlating a pixel location of the first image data set and a pixel location of the second image data set for which the image data match; and
compute an essential matrix that characterizes relative positions of the single camera at the first and second times according to the feature point correspondences.
20. The non-transitory computer readable medium of claim 18 , further comprising computer executable instructions to cause the processor to:
compute the first and second rectified image data sets for only the regions of interest that exclude the epipole.