Extraction of structured light stripes center from Non-uniform scattering surfaces fused with deep learning

The non-uniform strongly scattering surface enormously influences the measurement accuracy of high-precision structured light Three-dimensional (3D) surface topography. Accurately extracting the central profile of non-uniform highly scattering surfaces is the basis for 3D measurement using point cloud data. In this paper, based on the analysis of the properties of structured light stripes on non-uniform scattering surfaces, a high-precision subpixel laser stripes center robust detection method is proposed. This method first used the Stripe-Segmentor network to accurately segment the Region of Interest (ROI) of the linear structured light stripes, then used the gradient descent method to determine the normal direction of the center of the ROI area light stripes, and finally used the gray-scale centroid method to realize the sub-pixel in the center of the linear structured light stripes precision extraction. Through validation, it is obtained that the laser stripes center extraction algorithm proposed in this study has good accuracy and real-time performance, and has certain engineering value.