Approach of Camera Relative Pose Estimation Based on Epipolar Geometry

摘要

Estimation of camera relative positions is very important in computer vision. Assuming that the camera has been calibrated, this study presents two algorithms of camera relative pose estimation which one is to recover the camera motion parameters by essential matrix and the other by epipolar geometry. Algorithm 1 is commonly used in three-dimensional (3D) reconstruction. In algorithm 2, firstly linearize coplanar condition equation through Taylor's expansion and solve elements of relative orientation by iteration procedure, finally recover camera motion parameters. Compared with the above two algorithms, algorithm 1 requires at least eight pairs of matching points and consider degenerate condition, algorithm 2 simply requires five pairs of matching points. Experimental data reveals that the two algorithms are feasible and the results of them are close which have the higher robustness and can satisfy the requirement of 3D reconstruction.