The Research On Image-Based Virtual Environment Modeling Technology

With the further development of computer vision technology, the research of virtual scene drawing technology has become an important research part of virtual reality system. The research of traditional geometry-based virtual scene modeling methods is mainly depends on computer graphics, its main features include:the interaction between the user and the virtual objects is convenient, the depth information of virtual objects can be direct obtained and the viewpoint is freely. On the other hand, as the research of this method is mainly depend on the geometric complexity of virtual scene, it is difficult to achieve in ordinary hardware plateform. The Image-based virtual scene drawing method use pixel as the basic rendering element, the rendering speed is independent with the scene complexity, it is only depend on the resolution of the image. As a result, the current research of virtual scene modeling methods is mainly focused in image-based modeling field, especially in the image-based rendering (IBR) and image-based modeling (IBM) techniques.Consider of high complexity of many plant body, the traditional modeling methods about these species are based on procedural methods, and there exist certain difficulties in interactive control over the plant model:if the parameters are too little, it is not enough to describe morphological characteristics of the plant body; on the other hand, if the parameters are too many, the model is hard to be rendered and comprehended. To certain extent, this problem can be solved through combining procedural method and image based rendering method. In second chapter, we use Bezier curve as the Axial Deformation control curve to generate variety trunks and branches. As for leaves generation, we use Bezier curves to describe shape of leaf midrib, and then embed the star texture image of leaf into the space of local frames along the midrib curve. At last, we use the rapid star texture matching to synthesize the special plant. Through combining the diversity of geometrical model and the diversity of image texture, the variety of plant bodys can be generated and the rapid rendering can also be realized. Based on the same plriciple, if we combine the natural wind model and rapid texture synthesis technology, the dynamic plant scene under wind field can be generated rapidly.In the third chapter, we discuss the rapid modeling problem of large scale virtual scene, reasonable decompose the total scene and present a modeling method apply to non-regular objects based on multi-viewpoint photo sequences. It can avoid many adverse factors exist in direct drawing process with 3D modeling software such as the long rendering cycle, difficulties in drawing and difficulties in applying 3D reconstruction method. It is a effective method. Using this method in 3D virtual scene, we can real-time track position of visitor's viewpoint, dynamic project corresponding photos and achieve better visual effect in practical application.The 3D reconstruction is an important branch in computer graphics and has extensive application. Using the image-based 3D reconstruction technology, we can extract textures from images and quickly reconstruct highly realistic 3D model. So it has received extensive attention.In the fourth chapter, we discuss the 3D reconstruction problem of regular objects. Firstly, the sufficient feature points are obtained through feature extracting and cross matching process. Secondly, we take covariance as the basis of similarity measure and use initial matching algorithms to all feature points to get the initial matching point set. As to the wrong matching points generated in matching process, we combine loose matching method and Lemds method to implement the matching process step-by-step and finally remove them. Thirdly, the space point reconstruction and surface reconstruction method are used to reconstruct the 3D model of target object, and at the same time, the realistic texture is extracted from the collected images. Finally, the 2D images is triangularized and then mapped into 3D space, the texture is added to the 3D model, the realistic exterior model is obtained.In the fifth chapter, we discuss the 3D reconstruction problem of irregular objects and mainly focuse on the 3D mesh surface reconstruction. The 3D surface point information is obtained from the image, and then triangularized to create the realistic 3D surface grid models. In order to meet the requirements of realistic effects and reconstruction accuracy, camera self-calibration is realized firstly, and the intrinsic and exterior parameters of camera is obtained through the foundation matrix and Kruppa equation computation. As the second step, the surface points searching algorithm of 3D object is given, the surface points are determined by judging the position relation between space points and image shadow and then the 3D surface points set is created. Because the 3D surface points are discrete, the discrete point cloud data need to be triangularized and grided. In this paper, the local projection algorithm is used to triangularize the point cloud data and then obtain the mesh surface model of target object.