Sparse disparity map estimation from stereo-pair images previously rectified, DEMO.m contains examples with some stereo pairs from Middlebury Stereo Evaluation, and KITTI 2015 disparity challenge.
Occlusions and missing disparities are labeled as NAN values, to add the possibility to generate a dense disparity map with other frameworks
For an RGB or grayscale stereo pair images, first, the magnitude gradient is obtained, then a Census transform is performed according to the selected method; for SAD and NCC, Census transform generates an image with a range of 2^24 possible values, for Hamming, Jaccard, and Mutual Information a binary Census Transform generates a volume with binary vectors for each pixel. Then a matching stage is performed with the selected method and finally, a match and disparity consistency check is performed.
The function uses GPU when is available.
See DEMO.m, for examples.
If this work is helpful to you, please cite this work.
Victor Gonzalez (2021). Sparse disparity map estimation (https://github.com/alx3416/Sparse_Disparity), GitHub. Retrieved .
Gonzalez-Huitron, Victor, et al. “Parallel Framework for Dense Disparity Map Estimation Using Hamming Distance.” Signal, Image and Video Processing, vol. 12, no. 2, Springer Science and Business Media LLC, Aug. 2017, pp. 231–38, doi:10.1007/s11760-017-1150-3.
Find the treasures in MATLAB Central and discover how the community can help you!Start Hunting!