面向对象的多视影像快速真正射纠正方法研究

Digital orthoimage has been changing the way we understand the world and turned traditional computer mapping into a powerful decision-making tool, give a great push to the public applications of geospatial information. Limited to the pinhole imaging principle and terrain fluctuation, existing orthoimage suffered the problems of projection error, occlusion and shadow. Based on the comprehensive analysis of the mapping relationship between 3D object features and 2D image features, as well as the geometric and radiation characeristics, this project proposes a new methodology of object-oriented true orthoimage rectification based on multi-view imagery, which is designed to overcome the difficulties of existing pixel-oriented methods. The main research contents include： 1) the object definition and automatic extraction in image space and physical space, 2) global visibility index of object information, 3) high performance CPU/GPU co-processing, 4) prototype system and application experiments. By means of the combined rectification of geometry and radiation, not only the accurate geometric features can be kept, but also more complete texture structure can be preserved due to the self-adaptive consideration of shadow and occlusion. This research would provide great potential of new theory and innovation for automatical and intelligent true otho-image processing of multi-platform, multi-sensor and multi-view high resolution image data.

数字正射影像已经大大改变人们认识世界的方式，并将传统地图转变为强有力的决策工具，极大地推动了地球空间信息的大众化与社会化应用。受针孔相机成像原理和地形地物起伏的制约，传统正射影像在城市地区存在严重的投影误差、遮挡和阴影等问题。本项目在深入研究三维目标特征与多角度二维影像特征之间的映射关系，并综合分析高分辨率影像的几何与辐射特性基础上，系统提出了面向对象多视影像真正射纠正的方法，旨在突破传统“像素级”正射影像纠正的局限，主要内容包括：1）物方和像方对象的定义及其自动提取，2）物方和像方对象信息的全局可见性索引，3）CPU与GPU协同的高性能处理，4）原型系统与应用实验。通过基于对象的纠正，既能保持准确的几何特征，还能自适应处理阴影和遮挡问题以保持正射影像完整的纹理结构。为多平台、多传感器、多角度高分辨率影像数据的自动化智能化真正射纠正提供新理论新方法。

数字正射影像已经大大改变人们认识世界的方式，并将传统地图转变为强有力的决策工具，极大地推动了地球空间信息的大众化与社会化应用。受针孔相机成像原理和地形地物起伏的制约，传统正射影像在城市地区存在严重的投影误差、遮挡和阴影等问题。本项目在深入研究三维目标特征与多角度二维影像特征之间的映射关系，并综合分析高分辨率影像的几何与辐射特性基础上，系统提出了面向对象多视影像真正射纠正的方法，旨在突破传统“像素级”正射影像纠正的局限，主要内容包括：1）物方和像方对象的定义及其自动提取，2）物方和像方对象信息的全局可见性索引，3）CPU与GPU协同的高性能处理，4）原型系统与应用实验。通过基于对象的纠正，既能保持准确的几何特征，还能自适应处理阴影和遮挡问题以保持正射影像完整的纹理结构。为多平台、多传感器、多角度高分辨率影像数据的自动化智能化真正射纠正提供新理论新方法。