基于视频流的实时摄影测量方法研究

Aiming at the new challenges of photogrammetry which is being faced with the online real-time data processing problem, this project makes a research of photogrammetric method based on video stream propagation, which will establish a theoretical and technical basis for online real-time photogrammetric applications. For the problem of low efficiency on image matching, a novel real-time matching method is researched, which rely on the guide of fast video stream propagation. This research will provide a basic algorithm for online real-time image matching. Given that traditional aerial triangulation algorithm is not so suited for real-time applications, a method of incremental aerial triangulation is proposed, which will balance the efficiency and precision of error adjustment in the process of aerial triangulation. To improve the efficiency of dense matching, a new progressive refinement dense matching method is researched. It utilizes known information to guide a relatively fast image matching process for new pixels, and then these matched pixels will in turn contribute to the pre-existing information via progressive refinement. This method will achieve a real-time and robust dense matching result. A kind of real-time photogrammetric data processing system that has characteristics of lightweight and high efficiency will be designed to achieve at acquiring data and processing data synchronously, and it will significantly give a system support for online real-time photogrammetry. The method researched in this project has the feature of high efficiency and high flexibility; photogrammetric system whose adaptiveness to unconventional environment will be largely improved which will definitely be more qualified for various real-time applications, and of greatly practical value in solving problems such as emergency response, remote detection, and military reconnaissance and so on.

针对摄影测量面临的在线实时数据处理新挑战，研究基于视频流传播的实时摄影测量方法，为摄影测量方法的在线实时性应用建立理论和技术基础。针对影像匹配效率问题，基于视频流快速传播的引导，研究影像实时匹配方法，为在线实时匹配提供算法基础。传统空三方法对实时性应用适应性不佳，研究一种增量式空三解算方法，可平衡空三的效率和平差精度。为提高密集匹配效率，研究一种逐级加密密集匹配方法，使用已知信息引导新的像素进行快速匹配并加密已有数据，可获得实时稳健的匹配结果。设计一种实时摄影测量数据处理系统，能够在数据采集过程中实现数据处理，具有轻巧、高效的特点，为在线实时摄影测量提供系统支撑。本项目所研究方法具有实时性强、灵活性高的特点，能够提高摄影测量系统在非常规环境下的适应性，适用于多种实时性应用场合，对解决应急响应、远程探测、军事侦察等方面的问题具有实际应用价值。

当前，各类在线应用对传统摄影测量技术提出了新的要求，急需加强在线摄影测量相关理论和技术基础的研究，本项目针对摄影测量中的影像匹配和位姿解算等方面展开了以下研究：（1）研究了一种基于视频流传播的在线影像匹配方法，以特征点连续跟踪的方式替代传统的影像匹配方法，为匹配提供了较好的初始对应关系，提高了影像匹配的实时性；（2）针对在线影像位姿解算问题，研究了一种增量式空三解算方法，通过影像的初始位姿计算和关联可视区域内影像组的局部平差解算，可快速解算当前影像的位姿，该方法通过限定平差的影像数量实现了效率和精度的平衡；（3）建立了一种在线全景图拼接方法，验证了影像位姿的正确性，同时生成了一种带有几何信息的拼接全景图，可为一些应急性应用提供参考成果；（4）研制了一种在线摄影测量数据处理原型系统，充分考虑体积、重量、功耗等因素，形成了一套轻巧高效的嵌入式系统，可搭载于轻小型无人机进行在线数据处理。项目所研究的相关成果能够促进摄影测量方法解决更多的非传统应用问题，提高在线数据处理的能力，满足各类实时性应用需求。