Registration of HRSI and LiDAR Point Clouds Based on Distance Transformation Model

  Objectives  High resolution satellite images (HRSI) can provide spectral characteristics observation information of ground objects at low cost and high frequency, while light detection and ranging(LiDAR) point clouds can provide fine geometric structure. The fusion of two kinds of data can realize complementary advantages, and further improve the accuracy and automation of ground object classification and information extraction. The realization of geometric registration with sub-pixel accuracy is the premise of two kinds of data fusion.

  Methods  A fast registration method based on line element distance transformation model is proposed. The point clouds are used as the control source, and the typical line elements such as building edges in the point clouds are projected into the image space through the initial rational polynomial coefficient(RPC) parameters of the satellite image, and the iterative closest point registration is carried out with the line elements in the satellite image, so as to achieve geometric registration by means of refining RPC parameters. In this method, the distance transformation model is used as the search table of the iterative closest point, which greatly improves the operation efficiency. Furthermore, the latest progressive robust solution strategy is adopted at the same time of the closest point iteration, so as to ensure the robustness of registration in the case of too much noise.Registration experiments are carried out on GeoEye-2 data, Gaofen-7 data, WorldView-3 data with LiDAR point clouds data.

  Results  The results prove that the proposed method can achieve a registration accuracy of 0.4—0.7 m on three kinds of images by using the ground control points which are accurately measured and the operator's internal control points as checkpoints.

  Conclusions  It is significantly better than the strategy of mapping point clouds to 2D image and registering through multi-mode matching.