基于四元数最少特征点的相机位姿估计  

Camera Pose Estimation Based on Quaternion from Least Feature Points

作　　者：侯培国[1] 赵梓建 宋涛[1,2] HOU Pei-guo;ZHAO Zi-jian;SONG Tao(School of Electrical Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China;Qinhuangdao Institute of Audiovisual Machinery,Qinhuangdao,Hebei 066004,China)

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004 [2]秦皇岛视听机械研究所,河北秦皇岛066004

出　　处：《计量学报》2020年第5期551-557,共7页Acta Metrologica Sinica

基　　金：国家自然科学基金(61601400)。

摘　　要：为解决相机位姿估计的问题,提出了一种基于四元数最少特征点的相机位姿估计算法。在相机拍摄的二维图像中检测并匹配特征点,根据特征点坐标与约束条件建立多项式系统,通过求解对应的矩阵方程来求解多项式系统。用四元数表示相机的旋转,避免了求解中相机旋转与平移相互纠缠的问题,当2个摄像机视图之间的平移为零时能够很好地进行位姿求解。求解中结合5点算法对求解原理进行了详细推导,并进行抗噪声测试。测试中随着匹配的特征点数增加,算法平均估计误差范围不超过2%。利用KITTI数据集测试算法的实用性,并与其他几种方法进行结果比较。结果显示,相同条件下算法将估计精度提高了24. 5%以上,体现出良好的估计精度。In order to solve the problem of camera pose estimation,a algorithm based on quaternion from least feature points was proposed. A polynomial system,established by constraints and feature points which were detected and matched in the two-dimensional image,was solved by corresponding matrix equation. Using quaternion to represent the rotation of the camera,which avoided the entanglement between rotation and translation in the process of solving. When the translation between two views is equal to zero,the problem can be solved well. The algorithm was deduced in detail by combining fivepoint algorithm and the anti-noise test was carried out. In the test,as the number of matched feature points increases,the average estimated error range of the algorithm does not exceed 2%. The KITTI dataset was used to test the usefulness of the algorithm and compare the results with several other methods. The results showed that under the same conditions,the algorithm improves the estimation accuracy by more than 24. 5%,which shows a good estimation accuracy.

关 键 词：计量学 相机位姿估计 四元数 5点算法 计算机视觉 

分 类 号：TB96[机械工程—光学工程]