Rigid-array constraints and matching-quality-weighted global bundle adjustment for large-scale 3D-DIC reconstruction

To address error accumulation caused by local sequential stitching in multi-step 3D-DIC reconstruction of large-scale complex surfaces, we propose a matching-quality-weighted global bundle adjustment framework for rigid multi-camera arrays. By exploiting the physical prior that the relative poses among cameras in the array remain constant, the method parameterizes the multi-camera system in a reduced-dimensional form and incorporates all scanning steps and all camera observations into a single global optimization problem. Observation weights are further constructed from the zero-mean normalized sum of squared differences (ZNSSD) produced by DIC matching to suppress the influence of low-quality matches on reconstruction. Experiments on a large-aperture parabolic antenna show that, compared with Kabsch stitching and unweighted global BA, the proposed method reduces the stitching consistency RMS in overlapping regions from 0.716 mm to 0.1060 mm, while achieving a mean reprojection error (MRE) of 0.045 pixels. Vacuum thermal deformation experiments from -80°C to 100°C further verify the robustness of the method under extreme conditions.