Disparity estimation device, disparity estimation method, and program

A disparity estimation method, comprising:

• (a) obtaining a first image captured by a first camera disposed in a first position and a second image captured by a second camera disposed in a second position;
• (b) calculating, for each of a plurality of first pixels of the first image, a first census feature amount based on a comparison between a pixel value of the first pixel and a pixel value of a neighboring pixel of the first pixel in the first image, and calculating, for each of a plurality of second pixels of the second image, a second census feature amount based on a comparison between a pixel value of the second pixel and a pixel value of a neighboring pixel of the second pixel in the second image;
• (c) calculating, for each of the plurality of first pixels, a difference degree corresponding to a disparity value in a position of each of at least one of the plurality of second pixels with respect to the first pixel using the first census feature amount and the second census feature amount of the at least one second pixel;
• (d) obtaining a first disparity value of each of the plurality of first pixels with integer pixel accuracy by selecting, using the difference degree, a disparity value corresponding to a lowest difference degree;
• (e) extracting, for each of the plurality of first pixels, a plurality of reference pixels located in positions corresponding to the first disparity value of the first pixel and an additional disparity value different than the first disparity value from the plurality of second pixels, calculating a plurality of sub-pixel evaluation values based on a relationship between (i) the pixel value of the first pixel and the pixel value of the neighboring pixel of the first pixel and (ii) a pixel value of each of the plurality of reference pixels and a pixel value of a neighboring pixel of the reference pixel, and estimating a second disparity value of the first pixel with sub-pixel accuracy by equiangular fitting in which the plurality of sub-pixel evaluation values are used;
• and
• (f) obtaining a per-pixel disparity value in a disparity image for the first image with respect to the second image using the second disparity value of each of the plurality of first pixels,
• wherein at least one of (a) to (f) is performed by at least one processor.

A non-transitory computer-readable recording medium for use in a computer, the recording medium having a computer program recorded thereon for causing the computer to execute:

• (a) obtaining a first image captured by a first camera disposed in a first position and a second image captured by a second camera disposed in a second position;
• (b) calculating, for each of a plurality of first pixels of the first image, a first census feature amount based on a comparison between a pixel value of the first pixel and a pixel value of a neighboring pixel of the first pixel in the first image, and calculating, for each of a plurality of second pixels of the second image, a second census feature amount based on a comparison between a pixel value of the second pixel and a pixel value of a neighboring pixel of the second pixel in the second image;
• (c) calculating, for each of the plurality of first pixels, a difference degree corresponding to a disparity value in a position of each of at least one of the plurality of second pixels with respect to the first pixel using the first census feature amount and the second census feature amount of the at least one second pixel;
• (d) obtaining a first disparity value of each of the plurality of first pixels with integer pixel accuracy by selecting, using the difference degree, a disparity value corresponding to a lowest difference degree;
• (e) extracting, for each of the plurality of first pixels, a plurality of reference pixels located in positions corresponding to the first disparity value of the first pixel and an additional disparity value different than the first disparity value from the plurality of second pixels, calculating a plurality of sub-pixel evaluation values based on a relationship between (i) the pixel value of the first pixel and the pixel value of the neighboring pixel of the first pixel and (ii) a pixel value of each of the plurality of reference pixels and a pixel value of a neighboring pixel of the reference pixel, and estimating a second disparity value of the first pixel with sub-pixel accuracy by equiangular fitting in which the plurality of sub-pixel evaluation values are used;
• and
• (f) obtaining a per-pixel disparity value in a disparity image for the first image with respect to the second image using the second disparity value of each of the plurality of first pixels.