| In response to surveying and mapping support tasks such as battlefield environment monitoring and natural disaster damage assessment,light and small UAVs can respond quickly,provide image data at the first time,and quickly produce three-dimensional geographic information of the target area.But the UAV has some disadvantages,such as poor attitude stability and weak ability of measuring pose parameters.In order to ensure the accuracy of results,the aerial photography process often requires good conditions,and a large number of ground control points arranged according to the rules,which invisibly limits the applicable occasions of UAV and increases the field workload.With the development of RTK/PPK and SFM technology,they provide solutions for high-quality and efficient processing of UAV images,but their adaptability and robustness still need to be further improved.This paper combines difference post-processing algorithm,photogrammetric error theory,multi view geometry theory and image registration method to study the high-precision image location without ground control points and the fast image splicing in special areas.The main work and innovations are as follows:?1?The traditional differential UAV system has high integration and limited accuracy of acquiring image camera coordinates,so it still cannot get rid of dependence on control points.Aiming at this problem,a set of differential GNSS UAV system is integrated by low-cost BD930 modules.By constructing the carrier phase double difference model of the fusion BDS/GPS/GLONASS satellite,the image line of exterior orientation elements are obtained with high precision after cubic spline function interpolation and real-time eccentricity correction combined with airborne IMU.Experiments show that applying the differential UAV system,the high fixed solution ratio and positioning accuracy of centimeter level under the condition of short baseline can be obtained.?2?Aiming at the problem that current image positioning algorithms are weak and require high image input conditions,a robust SFM positioning method is proposed.Firstly,the rotation matrix is linearized by Lie group and Lie algebra theory,its accuracy and reliability are improved by the L1-IRLS algorithm combined with L1-norm and weighted L2-norm in the iterative calculation process.Then,in the process of gross error elimination,the reconstruction strategy of image relation graph is proposed.It effectively avoids iteration termination caused by local weak correlation.Finally,the rotation parameters of all images are converted to GNSS camera coordinate system,and the high-precision image exterior orientation elements,camera calibration parameters and ground point coordinates are calculated by GNSS-assisted self calibration bundle adjustment under the restriction of best station weight.Experiments demonstrate that the more robust SFM positioning method improves the positioning accuracy and the effect of 3D point cloud reconstruction.When the value of ?XY is 0.05m and the value of ?H is 0.1m,the optimal solution of 3D target information can be obtained.?3?The UAV image stitching method based on feature matching is time-consuming,requires high image overlap and texture features.Aiming at this problem,a fast UAV image stitching method based on POS is proposed.Firstly,according to the attitude information,the images are divided into two groups:stable group and disturbance group.The key frames are selected by minimum spanning tree and breadth first search algorithm.Then,the reference image is determined from the stable group by defining the weighted topology under the condition that global cumulative registration error is minimal.Finally,the homography transformation matrix between adjacent images is deduced by POS data.All the geometric transformated images are mapped to the reference plane to complete the global transformation estimation.Experiments show that applying the stitching method can effectively reduce the accumulation of registration errors and greatly shorten the stitching time. |
PDF Full Text Request