Multi-view Matching Theories And Methods For Automatic DSM Extraction From Linear Array Images

Represented by such characteristics as large-scale production, high automation and super intelligence and so on, Aerial and Spatial Remote Sensing and Photogrammetry are stepping into a new era. Under such a background, Digital Surface Model (DSM), a basic geographical information product, has to be produced precisely, automatically and rapidly. Therefore, in order to extract DSM form liner array digital images, this paper dives systematically and deeply into such techniques as obtaining and preprocessing of matching images, multi-view matching models and strategies for both point elements and line elements, automatic DSM extraction process etc. The work completed and accomplishments achieved in this paper are given as follows.1. An image preprocessing process, which is comprised of geometric rectification, adaptive smoothing, Wallis filtering and wavelet image pyramids, is designed for L0 Level ADS40 three-line array digital images to provide high quality images for multi-view matching.2. Multi-view matching models for Point elements and linear array images are investigated and compared; quasi-epiloar line for linear array images is introduced. Then, the searching direction, checking methods and working process for the models are given. Researches prove that the models can synthesize redundant image information to avoid blind information area, and thus, to improve the successful rate of matching.3. Advantages of point element multi-view matching models on similar features and occluded features are analyzed. Matching difficulties of the models on discontinued features are deeply investigated, and an adaptive window extending algorithm is proposed correspondingly.4. Dominant Point model for Line elements is improved. Assisted by the geometric and gray attributes, line elements are matched indirectly from the matching results of dominant points.5. Strategies for point element matching models, such as Correlation Window warping, Self-adaptive Matching Window extension, Feature-based Y-parallax compensation, Quasi-epipolar Line Constrained Least Square matching and so on, are proposed. Principles and functions of the strategies are analyzed, and the integrating modes and working processes of the strategies with matching models are given. Experimental results show that the matching models and strategies can be combined to reliably match dense gird points.6. Correlation Measure strategies, Probability Relaxation checking strategies and Linear B Spline edge reconstruction strategies for line elements are analyzed. Through the gray attribute of line elements and its corresponding dominant points, the reliability of matching results is improved after the checking of matching results to reconstruct precise 3D edges. 7. Based on multi-view matching models and strategies for both point elements and line elements, an automatic DSM extraction process for linear array digital images is proposed. Then, some important techniques, such as DSM triangle and Grid DSM generation, are discussed. Experimental results show that the process can be used to extract dense DSM form liner array digital images.