Binocular vision 3D reconstruction based on semiglobal block matching and weighted least squares filtering

3D reconstruction has extensive applications, while the monocular system suffers from depth deficiency, and the multi-eye system features high complexity. Accordingly, this paper proposes a binocular 3D reconstruction based on semi-global block matching and weighted least squares filtering. The method targets the challenges of stereo matching and disparity calculation, solving the optimal disparity by minimizing the global energy function. The subsequent addition of the WLS filter improves SGBM's robustness to weak texture and edge regions, optimizes the image details, and achieves high-quality 3D reconstruction. We also utilize matching and depth errors as quantitative metrics to evaluate the results. Experimental results demonstrate that the employed methods achieve pixel-level accuracy, fulfilling the reconstruction precision requirements.