The Research On Performance Comparison And Improvement Of Dem Matching Algorithms Without Ground Control Ponit

Digital Elevation Model (DEM) matching is a procedure of transforming multi-source, multi-temporal and multi-scale DEM data into a unified coordinate system. DEM matching without Ground Control Point (GCP) means that computers can automatically extract topographic features or geometric information from DEM to implement DEM matching without support of ground control points. Compared with the traditional methods with GCPs, DEM matching without GCP has the merits of safety, efficiency, time-saving, labour-saving, low-cost, and wide-application etc. It has broad and potential applications in surveying and mapping, remote sensing, navigation, geographic information system and other fields involved in DEM absolute orientation, DEM alignment and multi-source or multi-temporal DEM data fusion. Therefore, DEM matching without GCP has attracted much attention from a lot of domestic and international scholars, and is also one of the hot research topics in the field of surveying and mapping.The algorithms of DEM matching without GCP, i.e. the Least Squares Surface Matching (LSSM) algorithms including Iterative Closest Point (ICP), Least Z-Difference (LZD) and Least Squares3D surface matching (LS3D), are comparatively studied by some emulated experiments in this dissertation, to approach rather algrithms in the matching accuracy, the convergence efficiency and the pull-in-range. The nearest neighbor searching method based on Grid Partition (GP), through improving the nearest neighbor searching methods of k-d tree and Boxing Structure (BS), is proposed to improve the matching efficiency of algorithms of DEM matching without GCP. Generally, the pull-in-range of LSSM algorithms was small and it was easy to drop into a false global optimal using the traditional DEM matching model and GA (TGA) to solve the7-parameter transformation model with the scaling factor. In order to overcome these shortcomings, a new model based on "Maximizing the Sum of Matching-Degree (MSMD)" was proposed and used GA (MGA) to calculate the parameters of DEM matching model. Another new method of DEM matching combined GA with least squares (MGA+LSSM) was proposed to improve the matching efficiency and the matching accuracy of MGA and to explore the global optimization algorithms of multi-scaling DEM matching without GCP.The simulated tests based on regularly and irregularly distributed DEM data imitating different types of topography were carried out. The experiment results have shown:(1) The three "point-to-point" ICP algorithms of7-parameter model were basically the same in pull-in-range, matching accuracy and overall trend of convergence. When dealing with the case that the range of the DEM to be matched was completely included in the referenced DEM, the three algorithms could get larger pull-in-range and higher matching accuracy than the case that the range of the DEM to be matched was partly included in the referenced DEM. The simper the topography was, the larger the pull-in-range was. The more continuous the topography was and the more apparent the topographic features were, the higher the matching accuracy was. The larger the point density was, the higher the matching accuracy was and the lower the convergence efficiency was and the smaller the pull-in-range was.(2) When dealing with the irregularly distributed DEM data, the algorithms of the "point-to-plane" ICP, the LZD and the LS3D show higher matching accuracy than that of dealing with the regularly distributed DEM data. Among them, for irregularly distributed DEM data, the LS3D algorithm was the most excellent one in matching accuracy and convergence efficiency, and the "point-to-plane" ICP algorithm had the largest pull-in-range. For regularly distributed DEM data, LZD algorithm was the most excellent one in pull-in-range, matching accuracy and convergence efficiency.(3) The nearest neighbor searching method of GP was faster than the other methods such as exhaustive search, k-d tree and BS, and no influence on the matching accuracy of LSSM algorithms.(4) The MGA matching algorithm can avoid the defection that the TGA algorithm is easy to drop into a false global optimal. The MGA had the merit of large pull-in-range, and the LSSM had the merits of high matching accuracy and fast convergence speed. The matching algorithm of MGA+LSSM integrated the merits of both MGA and LSSM, thus, becomes more robust and has larger pull-in-range than LSSM algorithms.A variety of algorithms of DEM matching without GCP were implemented on computers and it can provide the necessary technical foundation of simulated tests and data analysis. The summary of the performance characteristics among various DEM matching algorithms without GCP suggests a valuable exploration to find the more adaptive multi-scaling DEM matching without GCP. The research results would have helpful for multi-source, multi-temporal and multi-scale DEM matching and data fusion.