摘要
针对大部分基于单目相机的位姿估计方法只适用于平面场景的问题,提出了一种同时适用于平面场景和立体场景的单目位姿估计方法.首先,将所有图像与参考帧进行特征匹配;然后,提取每帧图像中都出现的特征点的像素坐标,并结合地平面约束和相机内参矩阵构建出观测矩阵;其次,对观测矩阵进行奇异值分解得到各帧图像的位姿估计,并利用约束矩阵解决奇异值分解不唯一问题;最后,利用光束平差法优化图像位姿,得到机器人位姿的最优估计.实验结果表明:该方法能准确对移动机器人进行位姿估计.
This paper deals with a kind of pose estimation method which can be applied in planar scene and stereo scene simultaneously.Firstly,reference frame is chosen and matched with all pictures.The pixel coordinates of the feature points which appear in every frame are chosen for further calculations.After a series of transformation,the observation matrix is formed base on pixel coordinates of feature points,ground plane constraints and the internal parameters of camera.After that,the singular value decomposition(SVD)method is used to get the relative position of mobile robot,in the meanwhile,the constraint matrix is used to solve the problem of the result of SVD is not unique.At last,bundle adjustment is used to optimize the factorization results.The experimental results demonstrate the proposed method can realize the right pose estimation of mobile robot.
出处
《浙江工业大学学报》
CAS
北大核心
2018年第2期132-136,共5页
Journal of Zhejiang University of Technology
基金
国家自然科学基金资助项目(61202203)
浙江工业大学控制科学与工程学科开放基金(20140808
20150710)的资助
关键词
位姿估计
立体场景
也平面约束
观测矩阵
奇异值分解
光束平差法
pose estimation
stereo scene
ground plane constraints
observation matrix
singular value decomposition
bundle adjustment