Automatic registration of optical aerial imagery to a LiDAR point cloud for generation of city models
Automatic registration of optical aerial imagery to a LiDAR point cloud for generation of city models
复制标题
DOI：
10.1016/j.isprsjprs.2015.05.006
复制
发表时间：
2015-08-01
影响因子：
12.7
通讯作者：
Hardie, Russell C.
中科院分区：
文献类型：
--
作者：
Abayowa, Bernard O.;Yilmaz, Alper;Hardie, Russell C.
本文提出了一种框架，用于将从倾斜航空图像中提取的光学和 3D 结构信息自动配准到光探测和测距 (LiDAR) 点云，而无需事先了解初始对准。该框架在配准参数的估计中采用从粗到细的策略。首先，使用最先进的 3D 重建算法从光学航空图像中提取密集的 3D 点云和相关的相关相机参数。接下来，从激光雷达和光学图像衍生的点云生成数字表面模型（DSM）。然后根据从 LiDAR 和光学图像衍生的 DSM 中提取的显着特征来计算粗略配准参数。使用迭代最近点（ICP）算法进一步细化配准参数，以最小化配准点云之间的全局误差。所提出方法的新颖之处在于从 DSM 计算显着特征，以及使用几何不变量与归一化互相关 (NCC) 匹配验证相结合来选择匹配显着特征。特征提取和匹配过程能够自动估计初始化精细配准过程所需的粗略配准参数。配准框架在模拟场景和在真实城市环境中获取的航空数据集上进行了测试。结果证明，当共存的初始配准参数不可用时，该框架能够将从航空图像中提取的光学和 3D 结构信息配准到 LiDAR 点云。由 Elsevier B.V. 代表国际摄影测量与遥感协会 (ISPRS) 出版。
This paper presents a framework for automatic registration of both the optical and 3D structural information extracted from oblique aerial imagery to a Light Detection and Ranging (LiDAR) point cloud without prior knowledge of an initial alignment. The framework employs a coarse to fine strategy in the estimation of the registration parameters. First, a dense 3D point cloud and the associated relative camera parameters are extracted from the optical aerial imagery using a state-of-the-art 3D reconstruction algorithm. Next, a digital surface model (DSM) is generated from both the LiDAR and the optical imagery-derived point clouds. Coarse registration parameters are then computed from salient features extracted from the LiDAR and optical imagery-derived DSMs. The registration parameters are further refined using the iterative closest point (ICP) algorithm to minimize global error between the registered point clouds. The novelty of the proposed approach is in the computation of salient features from the DSMs, and the selection of matching salient features using geometric invariants coupled with Normalized Cross Correlation (NCC) match validation. The feature extraction and matching process enables the automatic estimation of the coarse registration parameters required for initializing the fine registration process. The registration framework is tested on a simulated scene and aerial datasets acquired in real urban environments. Results demonstrates the robustness of the framework for registering optical and 3D structural information extracted from aerial imagery to a LiDAR point cloud, when co-existing initial registration parameters are unavailable. Published by Elsevier B.V. on behalf of International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS).