3D reconstruction of complex weld geometry based on adaptive sampling

In this paper, an adaptive sampling algorithm is proposed for reconstruction of a complex weld geometry based on the obtained 3D point cloud. Several pivotal samples for reconstructing the weld geometry are selected from the point cloud using a randomized strategy in the initial stage of the iterative algorithm. Based on the pivotal samples, the model is incrementally refined in the second stage by adaptive sampling of the neighbors using three depth comparison functions. Our simulation results using the proposed sampling algorithm demonstrate significant improvements in total samples drawn, number of iterations, and computation time while ensuring zero duplicate acceptance, as compared to random sampling approaches with and without replacement.