| According to the estimate of World Bank, Digital Construction for a city, a size of millions of urban population, can promote quadrupling economic level and realize "fourfold leap forward". Realizing Digital City is not only an effective approach for changing the mode of economic growth, but also it can promote fast and good development in politics、economy、culture、education、science and technology and military affairs, etc. Researched object in this paper is a street elevation that is an important part of work in Digital City. Street elevation is a significant part of street landscape, therefore how to carry out 3D visualization for street elevation plays an important role in 3D city modeling.Image data in this paper are got by ground 3S, namely vehicle (wheelchair) and hand held shooting. Currently, most of the 3D modeling manually use 3DMax,which has labor-intensive、inefficient and low degree of automation problems. In order to solve these problems, this paper presents a set of high automatic schemes, which can be illuminated as follows:1) Vanishing point analyzing:Vanishing point plays a significant role in getting image orientation elements and rectifying large obliquity angle image. Furthermore, there is scarcely research on vanishing point error in home and abroad. And especially there is no one researching on relative error ellipse of vanishing point error. This paper will deduce the relationship between vanishing point and image orientation elements from two aspects—vanishing point geometry and Generalized point photogrammetry. And this paper will introduce that different shooting angles have different effect to vanishing point adjustment model. Before computing vanishing point, straight-line corresponding to vanishing point needs to be grouped, and this paper presents a new method by "RANSAC"+"Condition adjustment with parameters". Through RANSAC to get initial value of vanishing point, vanishing point is computed by adjustment, which can eliminate observed line of a great residual error so as to perfect line grouping. Vanishing points errors and their relative error are evaluated by virtue of error ellipse theory. Moreover, Vanishing points precision influenced by shooting angles is researched and analyzed.2) Point and Line matching:In the point matching, current popular SIFT operator, feature point matching algorithm, as well as a novel matching idea of opposing and unifying are introduced. In the line matching, in order to resolve some issues in current line matching, such as degeneracy of epipolar line constraint、local limitation of affine transformation registration、feature line inconformity of stereopair、and unconscionable constraint of parallax difference etc, this paper proposes a matching algorithm based on relaxation and least squares image matching. This algorithm takes both independent observation and local integral restriction into account, which comply with objective law——opposing and unifying. Moreover, adjustment combining least squares matching with "Generalized Point Photogrammetry"theory is presented in order to improve precision of line matching and simultaneously achieve space line. Many-to-many relationship of straight line matching is also analyzed.3) Analysis of close-range large obliquity strip image:It consists of relative orientation、model connection and bundle block adjustment. Usually only two vanishing points can be detected in elevation image, which leads to lack restriction of vanishing point in Z orientation, be a degree of freedom in Z orientation, and imprecisely solve orientation elements. Thus, solution method based on Generalized Point bundle block adjustment combining space line constraint is proposed. After that, rectifying and auto-mosaiking large obliquity images are implemented.4) Auto-recognizing concavo-convex edge of street elevation:Due to the negative influence of noise and algorithm of extracting straight line, concavo-convex critical edging lines can not be extracted. Thus, how to recognize those not extracted concavo-convex critical edging lines leaves much to be desire. This paper suggests a novel algorithm to resolve this problem by following:Firstly, space lines cluster to classify coplanar lines. Secondly, utilizing space coplanar lines fits space planes. Thirdly, space planes intersect to get concavo-convex critical edging lines. Furthermore, many algorithm of fitting planes will be researched、analyzed and compared in order to get optimal fitting and recognizing effect.5) 3D modeling:Through auto-tracking roof、auto-recognizing concavo-convex edge of street elevation and manual selecting a point of ground boundary, texture profile corresponding to object space model can be determined. After calculating the scale between object space model and image texture,3D-elevation can be reconstructed on the base of plane.Above all, the purpose in this paper is how to auto-reconstruct close-range street elevation. And the degree of automation of 3D-reconstruction is raised by means of various theories. A scheme of high feasibility and high automation to reconstruct 3D-elevation is presented and implemented based on line and plane feature, which is also independent of model parameter、vectogram or assistant of space point cloud. |
PDF Full Text Request