基于模型的目标立体构建方法研究

发布时间：2018-05-07 09:16

本文选题：遥感 + 三维重建　；参考：《哈尔滨工业大学》2014年硕士论文

【摘要】：遥感技术在当代快速发展，军事上和民用上都体现了其重要性。通过不同的传感器平台，可以获得大量的遥感数据，这些数据在精确提取地面目标三维信息，对目标进行立体构建起着重要的作用。随着人们对立体构建的要求越来越高，多平台、多时相、多传感器获取的数据都存在各自的优缺点，通过信息的互补，可以弥补不同条件下的不足，使构建结果更加完整，信息得到最好的恢复。而对目标的构建比整个场景的构建会有更高的要求，目标顶面及边缘的信息恢复尤为重要。本文首先建立一个对三维信息进行解算的模块，这个模块可以为下一步的立体构建奠定一个很好的基础。从一个方面来考虑，主要是分析比较了不同模型下的特点及优缺点，研究各种不同条件对参数的影响。主要研究严格传感器模型和通用传感器模型及其改进，确定内定向参数和外定向参数求解的数学模型，进而完成三维信息的解算。另一方面，比较了不同会聚角对解算精度的影响，为传感器获取最佳数据时的姿态进行确定，减小误差，减小整个系统中的误差传递，，对以后的数据应用其重要作用。之后，重点研究基于典型目标拓扑模型的立体构建方法。考虑到三维信息的解算过程中，由于获取数据或匹配误差，可能导致有效解算点的数目有限。在这种情况下，建立典型目标模型库，将复杂的目标简单化，根据不同模型输入不同的参数，最终实现立体构建。目标主要集中在建筑物、油罐以及简单桥梁上。当获取的有效解算点较多时，则可建立顶面模型，利用曲面拟合完成三维信息的补充。同时验证了解算点数目及分布对解算精度有重要影响。最后，提出一个基于多重邻域滤波法的目标三维信息恢复方法。DSM数据的最优化在目标立体构建中起着重要的作用。由于匹配失败，无法确定各部分上的有效解算点。该算法主要侧重于目标而不是其他方法侧重的整个场景。目标比整个场景的有效信息更少一些，但是其边缘更难恢复，恢复的要求更高。该算法不同于一个插值，它可以平滑的恢复目标的边缘。我们可以根据分辨率建立一个网格，每个节点上都有数据。实验表明该算法可以大大提高三维信息的完整性。优化后完整性稳定在85%左右，具有很好的鲁棒性。
[Abstract]:Remote sensing technology in the contemporary rapid development, military and civilian reflects its importance. A large number of remote sensing data can be obtained through different sensor platforms. These data play an important role in accurately extracting 3D information of the ground target and constructing the target stereoscopically. With the higher and higher requirements of three-dimensional construction, the data obtained by multi-platform, multi-phase and multi-sensor have their own advantages and disadvantages. Through the complementary information, we can make up for the deficiency under different conditions and make the construction result more complete. The information is best restored. And the target construction will have higher requirements than the whole scene construction, the information recovery of the top and edge of the target is particularly important. In this paper, we first build a module to solve the three-dimensional information, which can lay a good foundation for the next three-dimensional construction. From a point of view, the characteristics, advantages and disadvantages of different models are analyzed and compared, and the effects of different conditions on parameters are studied. The strict sensor model and the universal sensor model and their improvement are studied in this paper. The mathematical model of solving the inner and outer orientation parameters is determined and the three-dimensional information is solved. On the other hand, the influence of different convergence angles on the accuracy of the solution is compared, which determines the attitude of the sensor to obtain the best data, reduces the error and reduces the error transfer in the whole system, and plays an important role in the application of the data in the future. After that, the stereoscopic construction method based on typical object topology model is studied. In the process of solving 3D information, the number of effective solution points may be limited due to the acquisition of data or the matching error. In this case, the typical target model base is established, the complex target is simplified, different parameters are input according to different models, and finally the three-dimensional construction is realized. The goal is to focus on buildings, oil tanks and simple bridges. When the number of effective solution points obtained is more, the top surface model can be established, and the 3D information can be supplemented by surface fitting. At the same time, it is verified that the number and distribution of the solution points have an important effect on the accuracy of the solution. Finally, we propose a method of 3D information restoration based on multiple neighborhood filtering. The optimization of DSM data plays an important role in the stereo construction of target. Due to the matching failure, it is impossible to determine the effective solution points on each part. The algorithm mainly focuses on the whole scenario of the target rather than the other methods. The target has less valid information than the entire scene, but its edges are harder to recover and the recovery requirements are higher. This algorithm is different from an interpolation, it can smoothly restore the edge of the target. We can create a grid based on resolution, with data on each node. Experiments show that the algorithm can greatly improve the integrity of 3D information. After optimization, the integrity is stable at about 85%, and has good robustness.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】