The Precision Analysis of 3d Reconstruction Based on Generalized Stereopair

According to the idea of realizing the 3D reconstruction using different images obtained from different sensors, this paper focuses on the different-source high resolution remotely sensed imagery covering the same regions, constructs the generalized stereopairs, and investigates the space intersection mathematics models based on RFM. Then quickly extracts the ground 3D information, and realizes the 3D reconstruction of target objects. After that, this paper carries on the precision analysis towards the reconstructed three dimensional models, and compares the relative precision of these 3D models reconstructed utilizing three different methods, and then validates that the generalized stereopair can realize the 3D reconstruction of ground objects, and the precision can absolutely satisfy the requirement of production. 1. INTRODUCE Along with the development of satellite remote sensing (ab. as RS) technology, the high resolution RS imagery play a more and more important role in military, economy and government decision-making, etc. It becomes a hot researching point to realize targets 3D reconstruction using RS imagery. For example, in the military applications, it is a urgent need to obtain the medium and small scale terrain data and 3D model reconstruction of the whole world key military targets using the satellite RS imagery; In the development of city navigation system, it needs to complete the destination navigation and localization, the optimal route selection after the 3D reconstruction of cities or regions; In the research of city emergency system, it urgently needs to construct the 3D reconstruction model, and provide the data support for government decision-making in the forecast of disaster scope, the disaster effect evaluation, and the crowd dispersal. In photogrammtry and RS, the 3D reconstruction is familiarly obtained by stereopairs, and we can get many types of RS images for now. However, we still suffer from the following two critical reasons: (1) Many regions may have no stereopairs; (2) Even have, but very expensive. Investigating the reasons, there are some factors to cause them (Weixi Wang, 2007a): (1) Polity (information blockage to non-allies or regions); (2)Price (the prince of stereo RS imagery is usually more expensive than single image several times);(3) Data lacking (many special regions or targets usually have no stereopairs of same satellite); (4)Security (the high resolution stereopairs of some special regions are not provided). Fortunately, more and more imagery could cover the same area, which are of different satellite sensors or aerial cameras, these images therefore have different space resolutions and different colors. Focusing on the above two problems, and according to the idea of realizing the 3D reconstruction using different images obtained from different sensors, this paper proposes a method to construct a generalized stereopair taking the overlap high resolution images from different sensors or cameras as the left and right images, and establishes new rigorous space intersection model based on Rational Function Model (ab. as RFM), Then quickly extracts the ground 3D information, and realizes the 3D reconstruction of target objects. After that, this paper carries on the precision analysis towards the reconstructed three dimensional models, and compares the relative precision of these 3D models reconstructed utilizing three different methods, and then validates that the generalized stereopairs can realize the 3D reconstruction of ground objects, and the precision can absolutely satisfy the requirement of production. 2. CONSTRUCTION OF GENERALIZED STEREOPAIR BASED ON RFM 2.1 Principle The orientation model of image decides which kind of models should be used in space intersection. The common orientation models have the following types: Polynomial Model (PM), Collinearity Equation Model (CEM), Affine Model (AM), Direct Linear Transformation (DLT), and Rational Function Model (RFM). Among them, RFM has already become the main orientation model for high resolution imagery. Comparing with other models, the RFM has following advantages (Weixi Wang, 2007b; Wang Weixi, 2007c): (1) independence from sensors or cameras; (2) suit for any kind of coordinate system, and free form the coordinate system; (3) can imitate rigorous models of any existing sensors or cameras. According to the above advantages, this paper takes the RFM as the basis of generalized