基于二维图像序列的三维地理场景重建方法研究

发布时间：2018-05-13 12:04

本文选题：点云 + 特征点　；参考：《重庆师范大学》2013年硕士论文

【摘要】：随着地理信息产业的快速发展，实施数字城市乃至数字地球战略是信息化建设的重要组成部分，其中三维空间数据是建立数字化地球与智慧型城市的基础与前提。为了提高三维空间数据采集效率，人们发明了三维激光扫描仪来快速获取现实场景的三维点云数据，但也存在一些不足；为了解决激光扫描仪三维建模以及软件三维建模方面存在的问题，并使得三维场景能够具有绘制速度快、低成本以及真实感效果，人们提出了基于图像的三维重建技术，由于输入数据来源于现实场景二维图像，因此降低了三维重建的复杂度，建模效率较高，但如何进一步提高三维建模精度仍然值得我们继续研究。采集二维图像序列是三维地理场景进行重建工作的基础，本研究采用四种不同的图像采集方式获取场景的二维图像，其中每个采集方式的场景适用范围与优劣不同。首先，对获取的二维图像序列进行了预处理，包括图像增强与图像滤波处理两方面，使用了MATLAB的编程实现了图像的灰度转换与均值滤波；其次，采用了Harris特征点提取算法对图像中的特征点进行了提取，并分别在图像全景和图像局部区域两种情形下进行了特征点之间的匹配，对匹配效果中的错误匹配与正确匹配进行了量化值分析；然后，从基于双目视觉的三维点坐标估计角度，分别对双目横向模式、角度扫描模式和双目纵向模式的三维坐标计算进行了分析，再将双目视觉形式获取三维点坐标扩展至多视点形式获取三维点云，，多视点形式有效降低了图像特征点匹配的不确定性，消除或减少图像匹配引起的误匹配；再次，对生成的原始点云进行优化处理，使用了图像的预先掩膜处理与立体点云框定优化处理。本研究采用了三维点云的Delaunay三角剖分方法，得出了三维三角网格模型，并对三角网格模型进行了真实感纹理绘制，重建出了较为逼真的三维地理场景；最后，对不同图像采集方式重建出的三维地理场景进行了研究总结与展望。
[Abstract]:With the rapid development of geographic information industry, the implementation of digital city and even digital earth strategy is an important part of information construction, in which three-dimensional spatial data is the foundation and premise for the establishment of digital earth and intelligent city. In order to improve the efficiency of 3D spatial data acquisition, 3D laser scanner has been invented to quickly acquire 3D point cloud data of real scene, but there are some shortcomings. In order to solve the problems in 3D modeling of laser scanner and 3D modeling of software, and to make 3D scene rendering fast, low cost and realistic effect, 3D reconstruction technology based on image is proposed. Since the input data is derived from the 2D images of the real scene, the complexity of 3D reconstruction is reduced and the modeling efficiency is high. However, how to further improve the accuracy of 3D modeling is still worthy of further study. The acquisition of 2D image sequences is the basis of 3D geographic scene reconstruction. In this study, four different image acquisition methods are used to obtain 2D images of the scene, in which the scope of application and the advantages and disadvantages of each acquisition method are different. Firstly, the two dimensional image sequences obtained are preprocessed, including image enhancement and image filtering, and the gray level conversion and mean filtering are realized by MATLAB programming. Harris feature point extraction algorithm is used to extract the feature points in the image, and the matching between the feature points in the panoramic image and the local region of the image is carried out respectively. The quantization value of mismatch and correct matching in matching effect is analyzed. Then, from the angle of 3D point coordinate estimation based on binocular vision, the binocular transverse pattern is analyzed. The calculation of 3D coordinates of angle scanning mode and binocular longitudinal mode is analyzed, and then the binocular visual form of acquiring 3D point coordinates is extended to multi-view form to obtain 3D point cloud. Multi-view form effectively reduces the uncertainty of image feature point matching, eliminates or reduces the mismatch caused by image matching. Thirdly, the original point cloud is optimized. Image premask processing and stereo point cloud block optimization are used. In this study, the Delaunay triangulation method of 3D point cloud is adopted, the 3D triangular mesh model is obtained, and the realistic texture of the triangulated mesh model is drawn to reconstruct the realistic 3D geographic scene. The three-dimensional geographic scene reconstructed by different image acquisition methods is summarized and prospected.
【学位授予单位】：重庆师范大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P208;TP391.41

【参考文献】