| Currently, major geospatial projects such as digital city and smart city are promoted in our country, vehicle-borne mobile mapping technology (VMMT) plays an indispensable role in these major geospatial projects. Vehicle-borne mobile mapping technology characters in fast data capturing, high degree of automation, low cost etc., it is a new research zone of photogrammetry and remote sensing. The multiple view measurable images captured by vehicle-borne mobile mapping system can provide great production capacity, which provided a fast and efficient new way for GIS data collection.According to the development trends of VMMT, the academic research about VMMT has transferred from system integration to massive remote sensing data post processing. Because of the small format, big data capacity and complexity measure environment, now the street images captured by Vehicle-borne mobile mapping system are only used in visualizing and low precision measurement, they are hard to accomplish high precision measurement, high precision measurement of street images has been the bottle neck which restricted the use and development of VMMT, it is urgent to carry out researches in this area.This thesis focuses on the high precision and fast processing of sequential images captured by Vehicle-borne Mobile Mapping System, systematically researched the aerial triangulation problem of vehicle-borne street-level Image sequential, image sparse matching method, adjustment model and dense matching method for vehicle-borne sequential street-level images are proposed in this paper, the major research content include:(1) Research the vehicle under various constraints street-level image reliable sparse matching method. In order to obtain reliable than 3°of the same name point to SIFT features, for example, the distance to the nearest matching ratio measure, the integrated use of bidirectional constraints, epipolar constraint, disparity and gray side constraints and point in object constraints such as point reprojection error strategy obtain reliable corresponding points, then turn point by corresponding points, to meet the demand for aerial triangulation of the same name as the point.(2) Method to represent attitude using rotation vectors is proposed. Vehicle-borne sequential images are often of massive quantity, usually tens of thousands of images in a single project, and its exterior parameters are greater, if using traditional Euler angles to represent the attitude in aerial triangulation, the trigonometric functions used in iteration will greatly decrease the efficiency. In this paper, block adjustment model with rotation vectors representing the attitude was established, by representing the attitude with rotation vectors, the coefficient matrix elements and iterations are done in linear system, which get rid of trigonometric functions and improve efficiency greatly.(3) Method to analyze the characteristics of POS by block adjustment combined with control images is proposed. The initial images were set as control images whose exterior parameters were treated as known, then model the characteristics of POS error by analyzing the residuals, experiments confirmed the correctness of the method.(4) Weighting method according to the spatial resolution of images is proposed. Because of the special mounting relationship between vehicle-borne cameras, the images captured are often tilt according to the building facet, so the spatial resolution of building facet in images differs greatly, a weighting method according to the spatial resolution of images is proposed to decrease the influence of images itself to the positioning result.(5) Preprocessing conjugate gradient sparse method is brought into aerial triangulation and construction method using diagonal square root matrix corresponding to normal equation coefficient matrix as preprocessing matrix. Because of the great difference of scales between matrix eigenvalues, the coefficient matrix of large area aerial triangulation is nearly singularity, made it hard to convergent in aerial triangulation, the method proposed in this paper can improve the eigenvalue distribution in some content, and can effectively improve the stability and quality of aerial triangulation.(6) Facet (building facet and ground) image dense matching method based on SIFT is proposed. Provided that reliable corresponding image points have acquired during sparse image matching, firstly, fitting the facet feature by homographic constraint, secondly, rejecting wrong match results by gradient of disparity constraint, lastly, constructing Delaunay triangles using the reliable corresponding points, under the triangle constraint, accomplish image dense matching in the facet area. Experiments show that this method can significantly increase the number of corresponding points. |
PDF Full Text Request