Enhancing Point Clouds Accuracy of Small Baseline Images Based on Convex Optimization

Abstract

AbstractThis paper proposes a method for enhancing accuracy of point clouds which are generated from small baseline of sequence images. The main contributions are threefold: First, the constraints of image pair-wise are computed based on invariant feature. The correspondence problem is solved by iterative method which remove the outlier. To avoid the disadvantage of incremental structure from motion, the global rotation of cameras are estimated by a robust method in the second step. These global rotations are fed to the point clouds generation procedure in next (third) step. In contrast with bundle adjustment which can gain local minima of back-projection error in L2-norm, the proposed method utilized error minimization in L ∞ -norm to triangulate accurately 3D points recast in quasiconvex optimization form. The simulation results will demonstrate the accuracy of this method from large view scene images in outdoor environment.KeywordsSIFTcorrespondenceRANSACglobal rotation estimationconvex optimization