Improving Least-Squares Stereo Matching by Using Different Reference Points

A dense image matching technique for high-quality point cloud generation has been proposed. The proposed algorithm identifies the corresponding pixel at the surface discontinuities, which is challenging for image matching algorithms. The algorithm uses different windows with reference points at the...

Full description

Saved in:
Bibliographic Details
Published inJournal of the Indian Society of Remote Sensing Vol. 51; no. 6; pp. 1227 - 1239
Main Author Sadeq, Haval Abdul-Jabbar
Format Journal Article
LanguageEnglish
Published New Delhi Springer India 01.06.2023
Springer Nature B.V
Subjects
Algorithms
Corners
Covariance matrix
Cross correlation
Discontinuity
Dynamic programming
Earth and Environmental Science
Earth Sciences
Image quality
Least squares
Pixels
Remote Sensing/Photogrammetry
Research Article
Template matching
Urban areas
Surface discontinuities
Deep learning
3D point extraction
Digital surface model
Dynamic programming
Stereo imagery
Dense image matching
Online AccessGet full text

Cover

More Information
Summary:A dense image matching technique for high-quality point cloud generation has been proposed. The proposed algorithm identifies the corresponding pixel at the surface discontinuities, which is challenging for image matching algorithms. The algorithm uses different windows with reference points at the corners in addition to that at the centres. At each pixel, the template at the centre and at the corners has been tested, and the successful matching of the template has been selected using the least value of the variance–covariance matrix, which is measured by least-square matching (LSM). A technique for finding the approximate point is introduced also. It is very essential in the LSM to search within a few pixels to make the solution converge to find the corresponding point, which is very problematic at high relief areas. This problem is solved using dynamic programming. Analysis results showed that these techniques can extract the corresponding pixels at the surface discontinuities and identify high-rise surfaces of the buildings more robustly in the urban area. In the evaluation, the algorithm has been compared with the point cloud result obtained from semi-global matching, deep learning stereo matching and normalized cross correlation (NCC) area-based matching technique.
ISSN:0255-660X
0974-3006
DOI:10.1007/s12524-023-01699-9