An improved stereo matching algorithm based on digital image correlation in 3D shape measurement
Boxing Qian, Hanfei Pan, Wei Shao, Youzhuo Li, Yu Wang · Measurement Science and Technology · 2025
Abstract Speckle structured light reconstruction can quickly obtain the point cloud of the measured surface. It is widely used in reverse design and product inspection. In the reconstruction process, stereo matching is the core of the whole algorithm and directly affects measurement accuracy and efficiency. In this paper, based on 3D digital image correlation, a fast and accurate measurement method is proposed. On the one hand, in the stage of integer-pixel search, with the judgment of grayscale deviation and epipolar geometry constraint, a large number of impossible candidate pixels are eliminated in advance. On the other hand, in the stage of sub-pixel matching, the iterative format of shape function in inverse compositional Gauss–Newton is simplified, avoiding the tedious matrix inversion in the iteration process. Thus, the computational efficiency of stereo matching is raised. Furthermore, two constraints on matching residual and reprojection error are set to remove some corresponding points with large deviations, then reconstruction accuracy is improved. The numerical simulation shows that the proposed stereo matching method has shortened the time by nearly ten times compared to that before improvement. Finally, the surfaces of a standard ball and a leaf are measured and reconstruction effect is well. The relative deviation of measuring standard ball is −0.42% the method can be applied to the rapid 3D measurement of dynamic objects.