利用稠密结构特征的航空影像与LiDAR数据配准

Registration of Aerial Images and LiDAR Data via Dense Structural Features

针对航空影像与激光雷达点云数据间存在显著的几何畸变和辐射差异导致难以精确配准的问题,提出了一种基于几何约束和稠密结构特征的自动配准方法。该方法包括粗配准和精配准两个阶段。粗配准通过基于分块FAST算子的特征点提取和局部几何校正两个步骤,消除影像间明显的尺度和旋转差异。在精配准阶段,首先构建了一个结合一阶和二阶梯度信息的新描述符(second-and first-order channel features of orientated gradients,S-CFOG)来提取影像间稠密的结构特征,然后在频率域采用三维相位相关作为相似性度量进行同名点匹配。最后,利用同名点对外方位元素进行精化,实现对这两类数据的精配准。通过两组不同覆盖场景的数据进行实验,结果表明,该方法可达到1~2个像素的配准精度。

Aiming at the problem of difficult co-registration between aerial imagery and light detection and ranging(LiDAR)data due to the significant geometric distortions and intensity differences,an automatic registration method based on geometric constraints and dense structural features is proposed,which includes two stages:coarse and fine registration.In coarse registration,the obvious scale and rotation differences between images are eliminated by two steps:feature point extraction based on block-FAST operator and local geometric correction.In fine registration,a novel descriptor combining first-and second-order gradient information,named second-and first-order channel features of orientated gradients(S-CFOG)is first constructed to extract dense structure features of images,and then the three-dimensional phase correlation is used as the similarity measure to detect correspondences in the frequency domain.Finally,the obtained correspondences are not only employed to refine the exterior orientation parameters but also used to achieve the fine registration of the two types of data.Experiments are performed by two sets of aerial images and LiDAR data,and the results show that the proposed method achieves the registration accuracy within the range of one to two pixels.