估计多视点摄像机姿态的两步法  

A two-step algorithm for estimating postures of cameras located in different points of view

作　　者：向长波[1,2] 谢丹[3] 刘太辉[1,2] 宋建中[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,吉林长春130033 [2]中国科学院研究生院,北京100039 [3]湖南师范大学工学院,湖南长沙410081

出　　处：《光学精密工程》2008年第10期1982-1987,共6页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.60372062)

摘　　要：提出了一种估计多个视点位置上摄像机姿态的新算法,即两步估计法。该方法采用两幅图像的特征匹配点重建出物体的局部模型,以局部模型作为世界坐标系的中心,利用模型的三维特征点与图像特征点的对应关系,求解出每个视点位置上的摄像机姿态。模拟数据与真实图像的实验结果表明,在特征点定位误差<1 pixel的情况下,摄像机的平均方位误差<0.5°,平均位置误差<0.1 cm,模型重投影误差<1 pixel。A two-step algorithm to estimate the postures of cameras located in different points of view was proposed. The local 3D model of an object is reconstructed with matching points in two images. Then,by taking the local 3D models as the center of world coordinate system, the posture of every camera is estimated according to the correspondences between the 3D feature points and the image feature points of a object. The experimental results of synthetic data and real images show that the average orientation error, average location error and the re-projection model error of the camera are less than 0.5°,0. 1cm and 1 pixel, respectively, under the condition of the position error for feature point in less than 1.0 pixel.

关 键 词：摄像机 姿态估计 两步法 三维重建 三维测量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]