ABSTRACT In this article, a structure from motion (SfM) framework for oblique aerial images of man-made environments is proposed, covering the issues of determining overlapping images, feature extraction, image matching, rejection of erroneous correspondences, feature tracking, automatic transfer of ground control points (GCPs), and bundle block adjustment. One of the challenges that it is intended to solve is the reduction of the required manual work concerning the measurement of GCPs, in order to increase the degree of automation of the exterior orientation estimation process, through the usage of geometric constraints automatically imposed. Yet another challenge is the difficulty in matching correctly feature points among multiple oblique views that depict scenes with repetitive patterns and homogeneous textures. The proposed algorithm solves this by eliminating all erroneous tie points through the combination of multiple checks and geometric constraints imposed during the image matching procedure and a robust iterative bundle adjustment framework. The proposed SfM methodology is applied in different configurations of oblique images under non-ideal aerial triangulation scenarios characterized by lack of well-distributed GCPs as well as minimum manual image measurements. The results are analysed, focusing on the improvement of the accuracy of the exterior orientation parameters thanks to the proposed robust outlier removal technique as well as on the impact of the proposed scale-based weighting strategy for bundle adjustment of oblique images on the exterior orientation results. The proposed SfM framework proves to be a good alternative solution to existing commercial SfM methods.

中文翻译：

---

使用基于 SfM 的鲁棒束调整的倾斜航拍图像的外部方向估计

摘要在本文中，提出了一种人造环境倾斜航拍图像的运动结构（SfM）框架，涵盖确定重叠图像、特征提取、图像匹配、错误对应的拒绝、特征跟踪、自动传输的问题。地面控制点 (GCP) 和束块调整。它旨在解决的挑战之一是减少与 GCP 测量有关的所需手动工作，以便通过使用自动施加的几何约束来提高外部方向估计过程的自动化程度。另一个挑战是难以在描绘具有重复图案和同质纹理的场景的多个斜视图之间正确匹配特征点。所提出的算法通过将图像匹配过程中施加的多重检查和几何约束与稳健的迭代束调整框架相结合来消除所有错误的连接点来解决这个问题。所提出的 SfM 方法应用于非理想的空中三角测量场景下的倾斜图像的不同配置，其特征是缺乏分布良好的 GCP 以及最少的手动图像测量。对结果进行了分析，重点是由于提出的鲁棒异常值去除技术而提高了外方位参数的准确性，以及所提出的用于倾斜图像束调整的基于尺度的加权策略对外方位结果的影响.