基于匹配点分布模型的低空无人机影像可靠匹配方法研究

Low-altitude Unmanned Aerial Vehicles(UAV) image primarily refers those images obtained from light UAV flying at the height of 1,000 meters above the ground. These images are flexible for access, of fine timeliness. Compared with the remote sensing images acquired by sensors embedded on aerospace platform, low-altitude UAV images have the advantages of higher resolution and more viewing angles. One of the core techniques in processing UAV images is to obtain robust reliable evenly-distributed matching points. Aiming at the problem, this project studies the methods of robust reliable image matching to obtain evenly-distributed matching points. Our research will include three aspects. (1) Develop the mathematical model between the precision of UAV aerial triangulation and matching method, geometric constraints, the spatial distribution, accuracy, number of matching points, as well as image content. (2) Design the algorithm framework of equalizing matching point based on (1). (3) Study the robust method of gross detection. The project will improve the accuracy of image registration and auto-aerial triangulation of low-altitude UAV images by obtaining reliable even-distributed matching points.

低空无人机影像主要指来源于轻型无人机等飞行于地面上空1000米以下的平台传感器，获取灵活、时效性好。相对于航空航天平台传感器获取的遥感影像而言，低空无人机影像还具有分辨率更高、视角多等优点。无人机遥感影像空三处理精度的核心问题之一是获得分布均衡、稳健可靠的影像匹配点。本项目针对该问题，研究无人机影像均匀可靠匹配的方法和技术。研究内容包括：（1）建立匹配方法、匹配几何约束条件、匹配点的空间分布、数量、精度、影像信息与空三精度之间的数理统计模型；（2）设计基于匹配点均衡模型的无人机影像均衡化匹配算法框架；（3）研究快速可靠的粗差及噪声探测方法。本课题旨在通过探索使低空无人机影像匹配点的分布更加均匀，精度更加可靠，从而提高低空无人机影像配准、空三等几何处理的结果精度。

无人机遥感影像匹配是无人机遥感影像摄影测量几何处理的关键技术和基础问题，分布均衡、稳健可靠的影像匹配则是无人机遥感影像空三处理精度保障的核心。项目围绕影像均衡匹配开展研究工作，按计划逐步开展了匹配约束、匹配点的空间分布、数量、精度等与空三精度之间的统计关系研究；基于匹配点均衡模型的无人机影像匹配算法研究；快速可靠的粗差及噪声检测方法研究。项目探索了匹配点均衡模型；提出了面向均衡模型的匹配方法框架，提出了一种具有几何抗性的特征区域描述子，提出了一种基于投票线策略和仿射不变比率的特征匹配方法；针对稳健匹配的需求，提出了基于鲁棒性lq估计算子的粗差剔除算法，并进一步进行优化提出基于Geman-McClure函数加权的lq估计算子。项目进一步将研究成果应用至影像拼接和基于影像的三维重建等方面。