Registration Of Urban Airborne LiDAR Data And Aerial Images Constrained By Linear And Planar Structure Features

Compared with traditional aerial photogrammetry,Airborne Light Detecting and Ranging(Li DAR)technology has the advantages of high efficiency,high accuracy,strong penetrating power,and little weather influence.Spatial coordinates and spectral information of ground objects could be obtained by combining Li DAR data and aerial images,which has wide application prospects in the areas of automatic feature extraction and classification,change detection,and 3D modeling of cities.The premise of the combination of airborne Li DAR point cloud and aerial imagery is to integrate the two into a unified geographic reference frame.However,due to various errors,Li DAR data and aerial images can't achieve precise geometric alignment generally.Therefore,it is significant to research on high-precision registration methods of airborne Li DAR point cloud and aerial images in depth.Aiming to solving the problem that the registration accuracy of airborne Li DAR point cloud and image is greatly affected by the point cloud density,this paper used the linear and planar structure features distributed widely in urban areas for registration.The main work conducted in this paper are as follows:(1)Parametric representation of linear and planar structure features.Two typical linear feature representation methods were introduced.The junction structure was designed to express the planar features,and the basic projection geometric relationship of the junction structure was derived.Finally,a multi-view forward intersection model of the junction structure was given.(2)Matching of linear and planar structure features.The rough matching of the linear features of point cloud and images was performed according to geometric relationship of image space and object space,followed by matching results refinement based on the Random Sample Consensus(RANSAC)theory.The epipolar-geometrybased constraint and the vanishing-point-based constraint were used for matching of image junction.The region gray correlation was carried out based on the plane homography.Then accurate matching results were obtained by consistency detection and rematch.Finally,the method based on normal direction slicing detected Li DAR plane points corresponding to the junction structure,which provided control information for imagery bundle block adjustment.(3)Aerial imagery bundle block adjustment model under the constraints of linear and planar structure features was studied.Taking the distance between straight line segments extracted from aerial images and back-projected lines of the Li DAR linear features as the measure,the straight line features were introduced into the bundle block adjustment.Considering that the normal accuracy of the Li DAR plane is high,the coplanar constraint of junction structure and conjugate Li DAR points were employed to conduct block absolute orientation.Experimental results showed that the proposed method could achieve higher registration accuracy and overcame the influence of point cloud density on registration accuracy effectively.For multi-view images,even if the density of point cloud is low,both the horizontal and the vertical accuracy of the proposed method could reach 1?2 pixels of the aerial images,better than the accuracy level achieved by existing methods obviously.For conventional aerial images,the registration accuracy of the three methods used in the experiment reduced,and the proposed method has a slight ascendant in accuracy.