Research On Automatic Space Registration Of LiDAR Point Data And High Resolution Aerial Image

In recent years, with the development of the surface features observation techniques, the means of spatial information become diversified. Since the single way to obtain the spatial information is limited by the precision, the fusion of the multi-source spatial information has become a significance research topic. This paper presents the research on registering aerial image to Li DAR point cloud by the gradient mutual information of single image and the parallax information of stereo images. Proposed approach takes photogrammetry collinear equation as strict mathematic mode and three stages are involved:① Virtual "pinhole" imaging process restored from aerial image orientation parameters is made on urban Li DAR point clouds to generate simulated gray image-Li DAR depth image. Because generated Li DAR depth image is very approximate to aerial image in geometrical appearance, the registration difficulty arisen from distinct difference of spatial resolution, perspective distortion and size between two kinds of data sources can be alleviated greatly;② Using the gradient mutual information and parallax mutual information as similarity measurement, the geometry transform parameters between Li DAR depth image and aerial image are successfully estimated as well as image pyramid plus partitioning strategy is implemented to accelerate the parameter space search. In this stage, Li DAR laser feet points can be roughly mapped to aerial image pixels based on the estimated geometry transform parameters and the known projection relations between Li DAR point clouds and its depth image;③ Using all mapped aerial image pixels as observed values and the gradient mutual information or parallax information as weight, photogrammetry space resection algorithm is implemented to obtain improved image orientation parameters. All three stages are repeated until the given iteration calculation condition met and the registration of Li DAR point clouds with single aerial image realized.The main advantage of this approach is that the iterative gradient mutual information and parallax mutual information calculation successfully resolve the difficulty of heterologous correspondence between Li DAR point clouds and optical aerial image. In addition, with the help of intensive object information in Li DAR data, a good performance can be achieved with photogrammetry space resection algorithm. The experiments show that proposed approach is reliable, and with a high degree of automation and high precision.