基于无人机遥感图像的三维地表模型构建方法研究  

Study of 3D DSM Construction Method Based on UAV Remote Sensing Images

作　　者：吴俣[1,2] 余涛[1] 郑利娟[1] 谢东海[1] 

机构地区：[1]中国科学院遥感应用研究所遥感科学国家重点实验室,北京100101 [2]中国科学院研究生院,北京100049

出　　处：《系统仿真学报》2014年第2期376-381,共6页Journal of System Simulation

基　　金：高分重大专项项目(E0502/1112);中国科学院创新研究项目(Y35G1100CX);国产卫星遥感应用产品质量评价(Y3SZ380030)

摘　　要：近年来无人机遥感发展迅速,其高机动性、高分辨率和低成本等特点,已经被证明为低空遥感探测的重要手段。通过配准序列图像,最优化迭代得到内外方位元素,基于核线约束,利用改进的密集匹配方法,直接生成整个区域的数字地表模型。在密集匹配中,添加所有图像的FAST(Features from Accelerated Segment Test)角点,扩展投影点集,用于构建初始点云的格网平面。基于影像相关和核线约束,在相关的子区域内搜索最优投影点,然后通过最小二乘迭代各投影点,定位最优地面点。通过高斯函数的加权平均方法,引入各投影点的辐射比例因子,生成归一化的图像像素值,获得彩色点云。实验表明,该方法生成的真实彩色三维模型,有效地表现出建筑、道路和树木等基础地物类型。Recently, unmanned aerial vehicle （UAV） remote sensing is developed rapidly for high mobility, high resolution and low cost. This technology was proved as an important approach for low aerial remote sensing. Image registration and optimization iteration were used to obtain the elements of exterior and interior orientations. Based on epipolar constraint and the improved algorithm of dense matching, the digital surface model （DSM） of the area could be directly generated from the remote sensing images. In this algorithm of dense matching, the Features from Accelerated Segment Test （FAST） corner points were detected and expanded the projection pixels for the construction of the initial grid planes. Based on the image correlation and epipolar constraint, the optimal projection points were found out from the corresponding sub-areas, and these projection points were iterated using least square method for the optimal ground points. The colorized cloud points were obtained by the weighted average Gaussian function method, and the pixel values could be normalized by the radiation scaling factors of the projection points. In the UAV experiment, the true color DSM were obtained, which could be used to distinguish the buildings, roads and trees effectively.

关 键 词：无人机遥感 FAST角点 核线约束 密集匹配 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]