Research On Mutual Information Registration Of Airborne Lidar Point Cloud And Optical Image

More and more remote sensing data have being acquired from different platforms by various sensors with the rapid development of science and technology.Airborne LiDAR has being become a frontier remote sensing technology in the past two decades,being its characteristics of an active sensor,free of shows and occlusions and wide adaptiveness of weather conditions.However,because of its data lack of multispectral information,,it is difficult to directly obtain the semantic information of an observed object.Optical images,on the other hand,are lack of the three-dimensional information,but can provide rich spectral information,texture features and other semantic information.Therefore,airborne LiDAR and optical images can be combined in many applications like 3D modeling,geospatial data visualization,thematic mapping,to only mention a few.The registration of point cloud andoptical image is the premise for the fusion of the two datasets.How to perform a high-precision registration has been the focus of the study.Currently,though there are a lot of existing point cloud and image registration approaches,high precision and high level automation of registration is still a challenging task,especially when sufficient initial information is inadequate.Bearing the abovementioned issues in mind,the paper adopts mutual information as the basic measure to register point cloud and optical images,by converting point cloud into intensity image and depth image.The main research contents include:(1)Overview the state-of-the-art of the registration of airborne LiDAR point cloud and optical images and the development of mutual information for registration purpose.Introduction is also provided to basic issues of registration and the concept of mutual information and its calculation;(2)In order to improve the registration accuracy and the registration efficiency,initial parameters are needed.In the paper,the SIFT algorithm is used to firstly register the intensity images generated by the point cloud and optical images.According to the rough registration(Patches were extracted from both point cloud and images,then patch area was applied to determine the matched patches.The centers of gravity of the matched patches were used as corresponding points to establish a linear transformation model to finish the rough registration)to remove a large number of false matching point pairs.Two-dimensional and three-dimensional correspondence relation is established according to the elevation value of the depth image corresponding to the intensity image;(3)In order to make full use of the data,the point cloud is used to generate the intensity image and the depth image.Calculating the combined mutual information between the two images and the optical image.According to the maximum combined mutual information as a measure with Powell algorithm to complete the registration;(4)Three sets of airborne point cloud and corresponding optical image data in Jilin area were tested with the algorithm proposed in the paper.It is proved that the method of combining mutual information proposed in this paper is feasible and effective in term of the registration of point cloud and optical images through the qualitative and quantitative analysis of the results.