Relative pose estimation method based on single SIFT features
Position estimation is one of the fundamental problems in precision optical measurement and autonomous driving.For practical applications such as autonomous driving,where the camera moves on a plane and the degrees of freedom of the camera position are three,this paper proposes a camera relative pose estimation method based on a single SIFT feature.Since the monocular camera cannot recover the translation scale,the degrees of freedom of the camera motion are reduced to two degrees of freedom with only rotation and translation angles.By observing the ground,ground homography information containing camera motion and plane normal vectors can be obtained.Therefore,camera motion can be restored by extracting homonymous ground points to estimate the homography matrix.In order to reduce the number of RANSAC interactions and improve the efficiency of the algorithm,SIFT features are introduced to the pose estimation,which include the coordinates of the homonymous points in the two images,as well as their feature rotations and feature scales.So that the information contained in a single point pair can be expanded,and the number of point pairs required for solving the homography matrix can be reduced efficiently.For the case of planar two-degree-of-freedom motion,this paper uses a single SIFT feature point pair to complete the estimation of homography matrix,and then uses Random Sample Consensus algorithm to optimize the results,and finally decomposes the homography matrix to obtain the relative position estimation results.The proposed method is proved to be effective by comparing it with the 2pt method and 5pt method in simulation and real experiments.
NETL
NSTL
万方数据
