Disparity estimation for misaligned stereo image pairs
A disparity vector for a pixel in a right image corresponding to a pixel in a left image in a pair of stereo images is determined. The disparity vector is based on a horizontal disparity and a vertical disparity and the pair of stereo images is unrectified. First, a set of candidate horizontal disparities is determined. For each candidate horizontal disparity, a cost associated with a particular horizontal disparity and corresponding vertical disparities is determined. The vertical disparity associated with a first optimal cost is assigned to each candidate horizontal disparity, so that the candidate horizontal disparity and the vertical disparity yield a candidate disparity vector. Lastly, the candidate disparity vector with a second optimal cost is selected as the disparity vector of the pixel in the right image.
1. A method for determining a disparity vector for a pixel in a right image corresponding to a pixel in a left image in a pair of stereo images, wherein the disparity vector is based on a horizontal disparity and a vertical disparity and the pair of stereo images is unrectified, comprising the steps of:
determining a set of candidate horizontal disparities;
calculating, for each candidate horizontal disparity, a cost associate with a particular horizontal disparity and corresponding vertical disparities;
assigning the vertical disparity associated with a first optimal cost to each candidate horizontal disparity, wherein the candidate horizontal disparity and the vertical disparity yield a candidate disparity vector, and wherein the optimal cost is a least cost determined by a matching function C(x, y|i, j) for a horizontal disparity in a bounded horizontal range x −search ≦i≦+x search and for a vertical disparity in a bounded vertical range y −search ≦j≦y search , and wherein i is a horizontal offset and i is a vertical offset for a pixel in the image at a location (x, y); and
selecting the candidate disparity vector with a second optimal cost as the disparity vector of the pixel in the right imager, and wherein the second optimal cost is a least cost determined by the matching function C(x, y|i, j).
2. The method of claim 1 , wherein the cost is a least cost.
3. The method of claim 1 , further comprising:
performing a bounded two-dimensional search near the pixel.
4. The method of claim 1 , wherein the costs form a four-dimensional costs field.
5. The method of claim 4 , wherein the cost field is compressed to three dimensions.
6. The method of claim 3 , wherein the stereo images are converted to a stereo image pyramid, and the search is conducted from a coarse level to a fine level of the image pyramid.
7. The method of claim 6 , wherein a horizontal search at a particular level is performed around a scaled horizontal disparity calculated from a coarser level.
8. The method of claim 6 , wherein a vertical search at a particular level is performed around a scaled vertical disparity calculated from a coarser level.
9. The method of claim 6 , wherein a vertical search is performed for all of the candidate horizontal disparities.
10. The method of claim 8 , wherein a range of the vertical search is scaled according to the level of the pyramid.
11. The method of claim 8 , wherein the vertical search at each level is independent of the vertical search at other levels.