中文摘要：

为了解决相对定向过程中匹配得到定向点的合理定权问题,提高解算的精度和可靠性,提出了结合赫尔默特方差分量估计和选权迭代粗差剔除的相对定向方法.在格鲁伯区域,通过SURF（快速鲁棒）特征匹配和RANSAC（随机抽取一致性）算法剔除误匹配,提取大量亚像素级精度的相对定向点,利用赫尔默特方差分量估计法估算6个格鲁伯区域对应6组观测值的单位权中误差,并给其定权,然后使用选权迭代法进行粗差剔除,最终得到相对方位元素.利用相对方位元素校正后,航空影像和无人机影像上下视差最大值分别由传统方法的1.045像素和2.093像素减小到0.376像素和0.766像素;利用半全局匹配方法生成的视差图,明显优于传统方法生成的视差图.

英文摘要：

In order to determine reasonable weights for relative orientation points obtained by matching in relative orientation,and to improve the precision and reliability of relative orientation calculation,a method combining Helmert variance component estimation and gross error elimination using iteration method with variable weights was proposed.In Gruber areas,abundant relative orientation points with sub-pixel precision were extracted,by means of SURF（speeded up robust features）matching and RANSAC（random sample consensus）algorithm.Then,the mean square errors of unit weight and weights of 6sets of observed values corresponding to 6Gruber areas were calculated using Helmert variance component estimation method.Iteration method with variable weights was adopted to eliminate gross errors,and then relative orientation elements were acquired.The method was tested with aerial imageries and unmanned aerial vehicle low altitude imageries.Rectified by using relative orientation elements,the maximum vertical parallax in aerial imageries and UAV low altitude imageries are respectively reduced to 0.376 and 0.766 pixel from 1.045 and 2.093 pixel by traditional method,and the disparity map generated by semi global matching has been improved obviously.