Registration of Aerial Images and LiDAR Point Clouds by Exploiting Global–Local Geometric Constraints of Buildings

Reliable registration of aerial images and airborne LiDAR point clouds (ALS) is challenging during the integrated three-dimensional reconstruction of the two types of data. This study proposes a novel method for exploiting the global and local geometric constraints of buildings to automatically acquire reliable tie points between aerial images and ALS. First, dense matched point clouds (MPS) are obtained from aerial images. Building instances are extracted from the MPS and ALS using simple filtering and clustering algorithms, respectively. Owing to the similarity of the global geometric distribution of buildings in MPS and ALS, the relationships of building instances are established using a graph-matching method. Furthermore, to coarsely align MPS and ALS, line segments are extracted and matched with the local constraints of corresponding building instances. The proposed strategy can overcome the adverse effects of the significant differences between MPS and ALS in terms of density, completeness, and accuracy of the matching results. Moreover, the constraints of corresponding buildings greatly narrow the search range of line segment matching, improving the matching reliability and efficiency. Considering the local geometric deformations in MPS, the iterative closest point algorithm is used for each pair of buildings to realize precise registration. Subsequently, a strategy to select reliable tie points between aerial images and ALS is proposed for improving the accuracy of the orientation parameters of aerial images. The proposed method realizes the accurate registration of aerial images and ALS in two different scenes. The average projection errors of aerial images and ALS on two datasets are as low as 1.16 and 0.89 pixels, respectively.