Topology-Constrained Stereo Correspondence for Dot-Matrix Structured Light with Calibration-Disparity-Guided Local Search
Year: 2026
Type: article
Abstract: Binocular vision is widely used in structured-light three-dimensional measurement and industrial inspection. However, reliable stereo correspondence for repeated dot-matrix patterns remains challenging because neighboring dots are visually similar. To address this problem, this study proposes a topology-constrained binocular correspondence method integrating ordered grid localization with calibration-disparity-guided local search. An ordered lattice is constructed in the right view to provide refined dot locations and consistent spatial indices. A fixed disparity derived from stereo calibration is then used to restrict correspondence search to local regions in the rectified left view, reducing ambiguity among neighboring dots. Experiments on 20 stereo image pairs acquired from two stationary planar surfaces showed that the proposed method recovered all 121 lattice points in every acquisition, achieving 100% precision, recall, and F1 under a topology-derived reference-label protocol. Compared with a calibration-disparity global baseline using the same fixed disparity prior, the proposed method reduced the mean least-squares-plane residual standard deviation by 95.1% and 95.7% on the gray metal surface and white board, respectively, while reducing the mean residual peak-to-valley value by 94.5% on both surfaces. A controlled component study confirmed the complementary contributions of grid constraints and local association. Additional evaluations showed closer agreement with a screened coordinate measuring machine reference and complete lattice correspondence recovery at 50% linear image resolution. These results demonstrate that integrating ordered topology with calibration-guided local search improves stereo correspondence reliability and planar reconstruction consistency for repeated dot-matrix structured-light patterns.
Source: Optics
DOI: 10.3390/opt7060076
Institution: Shaanxi University of Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,448
OpenAlex ID: W7225818629
FWCI:
Cites:
Cited by:
Related to: