High-Fidelity 3D reconstruction via Global Gradient-Optimized speckle patterns and Color-Guided stereo matching with subpixel error Suppression

To address the challenges of high false matching ratios and low accuracy in stereo correspondence under single-frame speckle projection, this paper proposes a novel 3D measurement framework by integrating Global Linear Gradient Gaussian Speckle (GLGGS) with an Unrectified Color-guided PatchMatch Stereo algorithm (UC-PMS). This framework resolves the limitations of conventional methods in insufficient precision and high mismatching rates during 3D reconstruction. The GLGGS pattern employs a globally correlated grayscale gradient distribution and Gaussian-type speckle profile. It eliminates abrupt grayscale transitions at the profile edge while ensuring texture richness and pixel-level quantitative correlation, overcoming three major drawbacks of conventional speckle patterns: random distribution, background voids, and edge discontinuity. For subpixel matching error accumulation, the UC-PMS innovatively adopts a single-stage grayscale interpolation and a Census-HOG multi-feature fusion strategy, avoiding accuracy loss caused by cascaded calibration and interpolation in conventional algorithms. To validate the superiority of GLGGS&UC-PMS, simulations and experiments are conducted across multiple dimensions: Tilted plane reconstruction, depth measurement of tilted parallel planes, depth measurement of sphere-plane composites, 3D reconstruction of spherical objects. Results demonstrate that the proposed method maintains single-frame projection efficiency while resolving stair-like artifacts (point cloud discontinuities) in planar reconstruction, point cloud stratification/discontinuity in spherical reconstruction, and high false matching rates. It exhibits practical value for industrial applications, with its high precision and resolution meeting stringent requirements for depth measurement in precision manufacturing.