Unmanned Aerial Vehicle Images Aerial Triangulation Based On Compensation Model And Constrain Condition

Due to its portability, maneuverability and well-adaptive to environment, unmanned Aerial Vehicles(UAV) are more and more widely applied in the field of photogrammetry. The theories and experiments of UAV photogrammetric have become a hot issue in the field. The quality of Aerial triangulation directly affects on DEM, DOM and other products which make if very important in photogrammetry. The researches and experiments on GNSS supported photogrammetry with UVA proves that it can improve the quality of Aerial triangulation and reduce the work of measuring control points. It’s a significant research to process the UAV photogrammetry images while improve the quality of Aerial trigulation.The paper discusses the theories, methods and procedures of aerial triangulation UAV photogrammetry images in detail. Among them, we emphasized on the theories and experiments of increasing the quality Aerial triangulation, calculating the ground points coordinates in different measures, and try to built a compensation model, processed the images of JILIN city by Inpho software, and made experiments based on the MATLAB development platform. By doing these work, we get the following results:(1)By studying the least absolute、least square、least absolute-least square and least square constrained by the sum of residual error equal to zero, then using fours methods to aerial triangulation, analysis the theory precision and actual precision. The experiments turn to rectify that all of four measures used in aerial triangulation are practicable.(2)Built a compensation model to adjust errors of bundle block adjustment based on LS-SVM method’s theories and algorithm.(3)Processing aerial triangulation by Inhpo software, introducing the procedure, and built a compensation model to adjust the coordinates of the ground points using LS-SVM method which computed by bundle block adjustment. Besides, we test the methods of thin-plate spine interpolation and polynomial fitting to build compensation model. The results show that the method of LS-SVM have a better correct value in XYZ directions than the other two methods, indicating that we can use the LS-SVM to compensate the ground coordinates computed by bundle block adjustment.