Studies On Regional Extraction Techniques Of Primary Objects From Urban CCD Images

As one of the important information resources in 3D reconstruction of digital city, primary objects'regional extraction from urban CCD images is vital very much. This paper designs and realizes two fast algorithms of them with regarding building and road as two key goal areas in images. And different evaluation criterions of the two algorithms are provided for different purposes.It is very important to extract object region fastly and accurately in 3D reconstruction of digital city, and two key algorithms of this paper are basically designed for this acquirement. Extracting building region is mainly for reducing the scope of characteristic dot extracting and matching, improving the efficiency of 3D reconstruction. Secondly, it is used for setting up texture data base of buildings, and could realize texture mapping automatically. Besides the building extraction algorithm meets the requirement of time, it could also extract the building outline as accurately as possible. And it is knowned that the building region extracted is allowed to be larger slightly than itself range. Setting up road texture data base and realizing the texture mapping automatically is the key aim of extracting road region, so, the region extracted by road extracting algorithm should have less redundant information including trees, grass, etc..Two key algorithms of this paper are both based on linear features extraction. And this paper adopts the algorithm based on Hough transform, compared with another kind of linear features extraction algorithm. To solve the accumulating problem of classical Hough transform, Hough transform based on Dot-Density is proposed, which unites the space of images and Hough.Supposing models have been proposed before designing region extraction algorithm. The building extraction algorithm takes use of position information and geometry relations between contours in the building model, analyzes line arrays attained by Hough transform based on Dot-Density, and extracts the region of building. The road extraction algorithm utilizes the geometry between two boundaries and the yellow line in road model. They both use the thresholds of Dot-Density and Hough space counter to reduce the false line, while the thresholds are adaptive in extracting building algorithm. In addition, road extraction algorithm makes uses of mean value and variance of RGB value in colorful image to divide road into five areas, including road area, non-road area, etc., and Hough mapping occurs in area with edges of road only.This paper consists of six chapters. The first two chapters are introduction part and image enhancement, basic principle of boundary extraction and experiment verification part. The third chapter studies linear features extraction systematically, and the Hough based on Dot-Density is proposed. The fourth chapter and the fifth chapter are both key parts in this paper, and discuss model hypothesis, algorithm designing in building and road extraction separately. And the last chapter summarizes the whole work of this paper and the future plans.