Researches On Key Technologies About Three-dimensional Reconstruction Of Buildings Based On Omni-directional Image And Remote-sensing Image

Urban city is usually the political, economic and cultural center of the district. Digital city can not only be applied in civil fields, such as Intelligent Traffic System, City Management and Development Decision of the City, but also be used to construct the Virtual Battlefield Environment for military purpose. As an important section of the process of constructing the digital city, three-dimensional reconstruction of urban buildings has a wide range of application areas on both civilian and military fields.Catadioptric omni-directional image offers a large field of view which will reach to 360 degrees among the horizontal orientation. And one picture of this kind captures a large scene of the real world. In consideration of the superiority of the catadioptric omni-directional image and the shortage of other reconstruction methods, this paper proposes an approach for three-dimensional reconstruction of urban buildings based on catadioptric omni-directional image and remote-sensing image, and studies several key technologies about this problem.Firstly, the reconstruction based on catadioptric omni-directional image and remote-sensing image is divided into source image obtaining, image preprocessing, registration of catadioptric omni-directional image and remote-sensing image, extraction of building height, shape modeling of buildings and texture mapping. And we then acquire the processes of the reconstruction.Secondly, we propose two methods to register catadioptric omni-directional image and remote-sensing image. We prove one principle which gives the relation between angles determined by horizontal lines in those two kinds of pictures. The first method of registration is based on this angle principle, we first exploit building top-edges as primitive element, then search for feasible points among a point set via the angle principle, lastly we locate the best point from all feasible points based on statistical theory. There is a distance between one point and the optical axis of the camera captured the Omni-directional Image in the real world, and in the Omni-directional Image, there is also a distance between the image points of the point and the optical axis. We find out the relation between the two distances and design the second registering method based on it. The second method locate the best position the Omni-directional Image captured in by the angle relation defined by the angle principle and the vertical relationship defined by the distances relation, and then get the registration result. It has demonstrated on experiments that these two methods are all accurate and efficient for the registration, and the first method is more robust, the second method is faster.Thirdly, we give a algorithm for extracting building height. We abstract height extracting as geometry problems according to the model of catadioptric omni-directional imaging, then get the height by calculating the length of corresponding side. It has demonstrated on experiments that this method is accurate and efficient.