An improved 3D reconstruction method for weak texture objects combined with calibration and ICP registration

Abstract

3D reconstruction of weak texture objects is a major area of interest within the field of high-precision reconstruction. Because fewer feature points can be extracted from weak texture objects, the classical method of points cloud registration by extracting feature points is no longer feasible anymore, yet registration is an indispensable step for multiview 3D reconstruction. The classical method of point cloud registration, the ICP algorithm, requires less feature point extraction but depends on the initial iteration position. When the initial position is not suitable, the registration process becomes especially time-consuming and tends to degrade the registration accuracy by falling into the problem of local optima. Therefore, a novel method that combines calibration and ICP registration is developed to complete the reconstruction of weakly textured surfaces. By using the calibration results to roughly unify multiple point clouds under the same coordinate system, which gives a better initial position for subsequent ICP registration, to reduce the iterative time consumption problem of ICP algorithms. Experiments demonstrate the feasibility and transferability of the developed method and verify the effectiveness of solving the time-consuming problem of the ICP algorithm. Moreover, it can improve the accuracy of the 3D reconstruction method only by coordinate transformation. Hence, this method can be well used to complete the 3D reconstruction of weak texture objects.