Research On Aerial Triangulation Of UAV Images Assisted By Vehicle-borne Laser Point Cloud

With the emergence of new areas such as big data,smart cities,intelligent transportation,and unmanned vehicle,the acquisition of spatial information becomes more and more important.Aerial photogrammetry has been widely used as a relatively mature method to acquire spatial basic data.Especially the appearance of digital photogrammetry has greatly improved the automation degree and efficiency of production.At present,the theory of aerial photogrammetry technology has entered the automatic aerial triangulation phase.With the continuous improvement of GPS/IMU accuracy,air aerial triangulation has entered the phase of GPS/IMU assisted automatic aerial triangulation,which has reduced the need for control points,and has become a common technical measure for large-scale aerial image surveying.In recent years,new types of sensors have emerged in an endless stream.Research into the integration of multiple sensors has become a new research direction.The laser scanning system is another technical revolution following the appearance of GPS/IMU.Compared with optical sensors,it avoids the cumbersome process of image matching and aerial triangulation.It can quickly acquire three-dimensional spatial information on the surface of the object,while recording the intensity information of the point cloud.However,it still has the obvious defect that it cannot obtain rich texture information on the surface of the object.The laser point cloud(LiDAR point cloud)scanned by the laser scanning system has high-precision spatial 3D information.By using this feature,the LiDAR point cloud is used as a constraint to assist the aerial triangulation process,providing a new technical approach for less-controlled or uncontrolled aerial triangulation.In this paper,vehicle-borne laser point cloud and UAV images are taken as the research objects.By analyzing the research status of LiDAR point cloud-assisted aerial triangulation,a non-rigid method of vehicle-borne LiDAR point cloud-assisted aerial triangulation of UAV images is proposed,which greatly reduces the number of control points and improves the automation level.This method firstly uses the GPS/IMU as auxiliary information to establish the initial topological relationship of the image,improves the efficiency and quality of the automatic measurement of the image point,and uses the self-calibration bundle block adjustment to solve the image’s three-dimensional feature points in the geodetic coordinate system.Then,the obtained three-dimensional feature points of the image is used to register with the LiDAR point cloud.Finally,the registration parameters are introduced into aerial triangulation for iterative calculation to achieve the purpose of correcting the external orientation elements of the images.This method converts the LiDAR point cloud-assisted aerial triangulation into a registration problem between two cloud points.Aiming at the shortcomings of the current ICP point cloud rigid registration method,a non-rigid registration method between point clouds is proposed in this paper,and a new mathematical model is obtained by introducing the registration parameters into the bundle block adjustment.Using the cyclic iteration idea,the external orientation elements of the image are solved.Compared with the ICP algorithm,this method can solve the problem of non-rigid image distortion,the adjustment model is stricter,and the accuracy of the aerial triangulation is higher.The accuracy of the method is verified by comparing the coordinates of the measured checkpoints with the coordinates of the corresponding points obtained from the forward intersection using the external orientation elements obtained by this paper’s method.The results prove that this method is effective.