US 12,067,740 B2
Methods and apparatus for using filtering to improve image patch matching and/or for using depth information generated from stereo images
Atul Maharshi, South Orange, NJ (US); Pablo A Anigstein, Palo Alto, CA (US); and Rajiv Laroia, Far Hills, NJ (US)
Assigned to Deere & Company, Moline, IL (US)
Filed by Deere & Company, Moline, IL (US)
Filed on Jul. 21, 2021, as Appl. No. 17/382,335.
Claims priority of provisional application 63/055,167, filed on Jul. 22, 2020.
Prior Publication US 2022/0028101 A1, Jan. 27, 2022
Int. Cl. G06T 7/55 (2017.01); G06T 5/50 (2006.01); G06T 5/70 (2024.01); G06T 5/73 (2024.01)
CPC G06T 7/55 (2017.01) [G06T 5/50 (2013.01); G06T 5/70 (2024.01); G06T 5/73 (2024.01); G06T 2207/20024 (2013.01); G06T 2207/20224 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method, the method comprising:
operating a processor to perform, for a pixel location p in a reference image and a disparity d, the steps of:
i) determining, for the disparity d, a corresponding pixel location q in a second image corresponding to the pixel location p in a reference image;
ii) determining filter coefficients for filtering a candidate patch of the second image, for the disparity d, of said second image that includes said determined corresponding pixel location q;
iii) filtering a single one of: the candidate patch of the second image and a reference patch of the reference image, for the disparity d, using a filter implemented using the determined filter coefficients, to generate a filtered candidate patch for the disparity d, said single one of the candidate patch of the second image and the reference patch of the reference image which is filtered using said filter being the candidate patch of the second image, said filtering including:
blurring the candidate patch when the reference patch of the reference image is blurrier than the candidate patch and
sharpening the candidate patch when the reference patch of the reference image is sharper than the candidate patch;
iv) computing a matching cost value based on how well the filtered candidate patch for the disparity d matches the reference patch;
v) storing in memory the generated matching cost value for pixel location p and disparity d;
repeating steps i), ii), iii), iv) and v) for at least one other disparity d; and
determining a depth for pixel p based on the computed matching cost values corresponding to pixel p.