Registration Of Airborne LiDAR Point Cloud And Aerial Imagery For Desert Areas

In recent years, the observation mode of remote sensing is developing in a way of multi-sensor and multi-source data fusion. With the rapid development of hardware technology and the increasing need of the market, airborne laser scanning technique has become an important data acquisition method, which is complementary to the traditional aerial photogrammetry. Fusion of airborne lidar point clouds and aerial optical images can aquire the geometry, spectral and texture information of the observed surface quickly and directly, which overcomes the limitation of using only one sensor. Image registration is the prerequisite for multisource data fusion. Data fusion can be achived only if the data from different sensors are in a unified coordinate system.The registration of airborne lidar point clouds and aerial images has been studied mostly in urban areas. The various artificial ground objects in urban areas contain substantial semantic features that can be used as registration primitives for airborne lidar point clouds and optical images. Recently, sandstorms, springing from desert areas and influenced by humans, have become a serious problem to many countries. More and more communities focus their research interest on the deserts. However, traditional registration methods for urban areas have difficulties in texture-insufficient desert, because there aren't artificial ground objects which can be provided as registration primitives. Thus new registration primitives and registration strategies of airborne lidar point clouds and aerial images should be proposed in desert areas. This paper aims to solve the registration problem. Based on the registration, the airborne lidar point clouds and aerial images are fused to research the applications in desert areas. We focus on the classification of desert sand ground and bushes.The main contents of this study ares listed as below.(1) A registration method using centroids of bushes as control information is proposed. In accordance with the difficulties of registration in desert areas, this paper in the first time proposes a registration approach using the centroids of bushes as registration primitives. We found that there are many unevenly distributed bushes dotting the wild landscape which can be utilized for registration. The bushes are randomly distributed and any triangle connecting three of the bushes shows a unique geometric structure. Furthermore, the bushes in both the aerial image and the lidar data share the similar distribution, and are easily detectable in both data sources. In view of this, this approach employs similar triangles created from both centroids as the evidence for matching and verifies the registration by the RANSAC algorithm. A point matching algorithm using similar triangles is used to identify the right correspondences from a large number of possibilities. Finally, using matched centroids as control points, the airborne lidar point clouds and aerial images are successfully registered.(2) A registration method using sand ridges as control information is proposed. As for desert areas without bushes, the registration methods using centroids of bushes are no longer applicable. So this paper proposes another registration method using sand ridges as registration primitives. Firstly, sand-ridge information is extracted from both the aerial image and the lidar point cloud. Secondly, by extending the iterative closest point (ICP) approach, a perspective-ICP algorithm is proposed that achieves data registration through matching sand ridges. To automatically deal with outliers, an adaptive weighting strategy is adopted. Finally, with the weigthed perspective-ICP algorithm, the airborne lidar point clouds and aerial images are successfully registered.(3) An application based on fusing registered airborne lidar point clouds and aerial images are studied. After registration, the airborne lidar point cloud and the aerial image are in a unified coordinate system, making it possible to take the complementary advantages of these two data sources. The fused data provide basis information for estimation the desert topography and ecology. This paper focus on the classification of desert sand ground and bushes using fused airborne point clouds and aerial images. Combined geometry and spectral information, it is able to estimate the growth status of bushes. The classification accuracy is greatly imporved using fused data than only using airborne lidar point clouds, thus verified the importance of data registration and fusion.