集群环境下无人机影像快速拼接及点云生成技术研究

发布时间：2018-04-09 02:11

本文选题：并行处理　切入点：集群　出处：《解放军信息工程大学》2013年硕士论文

【摘要】：无人机遥感作为传统航测手段的重要补充，在很多领域得到了广泛应用，尤其在时效性要求高的应用中有着不可替代的作用。本文将集群并行计算技术应用于无人机影像的快速处理，对无人机影像拼接和三维点云生成技术进行了较系统的研究，选题具有理论意义和实用价值。论文的主要工作如下： 1、阐明了计算机集群的概念和特点，总结了集群的发展现状，详细介绍了两种典型集群式摄影测量系统的特点，在此基础上构建了基于HP刀片机的并行计算平台。 2、针对无人机影像航带连接点提取工作量大的问题，研究了重叠区域连接点提取的方法，并给出了一种多核并行处理方法，显著提高了处理速度。 3、在对SIFT特征匹配算法深入研究的基础上，设计了一种主从模式的序列影像SIFT匹配并行处理方法，提出了分层列划分的数据划分策略，并通过重叠通信和计算来减少通信开销，实验证明该方法可扩展性较好；主从模式方法在特征搜索匹配环节仍为串行处理，进一步对其改化，设计了一种对等模式的序列影像SIFT匹配并行算法，通过合理的任务分配和执行时序安排实现整个匹配流程的并行处理，实验结果表明，算法达到了接近线性的理想加速效果。 4、研究了解决影像拼接过程中辐射信息一致性问题的方法，设计了一种影像融合重构并行处理方法，在实现合成影像灰度过渡平滑自然、色调均衡的同时，提高了处理效率。 5、给出了一种基于无人机影像SIFT特征匹配生成三维点云的方案，在选定的参考坐标系中，，通过相对定向、前方交会得到首个模型，再通过“后方交会—前方交会—无控制点光束法平差”方法扩展几何模型，直至得到整个测区的三维点云。实验结果证明了该方法的可行性，为有控制点条件下进一步获得目标三维信息提供依据。
[Abstract]:UAV remote sensing, as an important supplement of traditional aerial survey methods, has been widely used in many fields, especially in the application of high timeliness requirements.In this paper, the cluster parallel computing technology is applied to the rapid processing of UAV images, and the technology of UAV image mosaic and 3D point cloud generation is systematically studied. The topic is of theoretical significance and practical value.The main work of the thesis is as follows:1. The concept and characteristics of computer cluster are expounded, the development status of cluster is summarized, the characteristics of two typical cluster photogrammetry systems are introduced in detail, and a parallel computing platform based on HP blade machine is constructed.2. Aiming at the problem of the heavy work of extracting the joint points in UAV images, this paper studies the method of extracting the joint points in overlapped regions, and presents a multi-core parallel processing method, which improves the processing speed significantly.3. Based on the in-depth study of SIFT feature matching algorithm, a master-slave pattern parallel processing method for SIFT matching in sequential images is designed, and a data partition strategy for hierarchical column partitioning is proposed, and the communication overhead is reduced by overlapping communication and computation.Experiments show that the method is extensible and the master-slave pattern method is still serial processing in feature search and matching, and a parallel algorithm of SIFT matching for sequential images with equal pattern is designed.The parallel processing of the whole matching process is realized by reasonable task allocation and execution scheduling. The experimental results show that the algorithm achieves an ideal acceleration effect close to linearity.4. The method to solve the problem of consistency of radiation information in image mosaic process is studied, and a parallel processing method of image fusion reconstruction is designed, which can realize the smooth and natural gray level transition of synthetic image and improve the processing efficiency at the same time.5. A scheme of generating 3D point cloud based on SIFT feature matching of UAV image is presented. In the selected reference coordinate system, the first model is obtained by relative orientation.Then the geometric model is extended by the method of "rear rendezvous-front rendezvous-no control point beam method adjustment" until the 3D point cloud of the whole measured area is obtained.The experimental results prove the feasibility of the method and provide the basis for further obtaining the 3D information of the target under the condition of control point.
【学位授予单位】：解放军信息工程大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P231

【参考文献】