DEM是国民经济和国防建设中广泛使用的基础地理空间信息，摄影测量方法是实现DEM全自动提取的有效而又经济的手段，其关键在于影像的密集匹配，一直是摄影测量学的研究热点。本项目遵循“航空摄影→GNSS辅助空中三角测量→影像密集匹配→地物点滤波→不规则三角网内插DEM”的严格摄影测量作业流程，深入研究各环节的数据处理技术，在自身多年研究已突破前两个和最后一个环节的瓶颈技术的基础上，着重研究三、四两个环节。拟针对航摄影像的特点和DEM提取需求，通过构建光流场拟合模型、局部能量最优函数和深度学习网络，提出一种基于光流场的影像密集匹配方法和一种基于深度学习的地物点滤波方法，设计一套能将其用于航摄影像的高精度DEM提取技术方案，以建立一种可真正实现全自动DEM提取的数字摄影测量新方法，通过开发实验原型系统及可行性实证研究，为大比例尺地形测图提供高效、高精度、高可靠性的全自动地形提取技术奠定理论和方法基础

Digital evaluation model (DEM) is one of the fundamental geospatial information widely used in national economic constructin and national defense construction. In recent decades, photogrammetric method has become the most effective and economical means for DEM automatic extraction. Meanwhile, dense image matching is the key to it, which is a hot topic in photogremmetry and computer vision field. In this project, a rigorous phtogremmetric workflow “GNSS controled aerial photography→GNSS supported triangulation→dense image matching→ground filtering→triangulated irregular networ (TIN) based DEM interpolation” will be followed. After years of research, we have already broken through the key problems of first two links and the last link, the comprehensive study of the third and fourth link will be focused in this project. Based on the characteristic of existing aerial image and the real application requirements.This project propose an optical flow field based dense image matching method and a deep learning based ground filter approach using optical flow fitting model, local optimization energy function, pixel tag based sample selection and hierachical deep convolutional neural network. And design a high precision DEM extraction scheme which can be used in realizing the actual automatic DEM extraction via digital photogrammetry techniques. Through the development of experimental prototype system and the feasibility experiments, it is expected to establish the theoretical and methodological basis for high efficiency, high accuracy, high reliability terrian extraction technology for large scale topographic mapping