Stereo Information Acquisition And Visualization Technology Studies Based On Near Space Platform

The space of20km to100km altitude is defined as near space, which isgenerally located between the highest flight altitude of aviation aircraft and thelowest orbit of satellite, and it is a relatively blank area of the aircraft. With theincrease of aerospace vehicle, the aerospace powers come to realize the potentialvalue of such a blank area in reconnaissance, surveillance and communication.Theinformation people get from the outside world is three-quarters visual. Comparedwith satellite images, images collected by near space platform are generally withhigher spatial resolution,wider angle, more freedom of time and space, and they canbe promising complements to satellite images and aerial images. As a newly-bornthing, near space platform has important strategic significance and broad applicationprospects, with which stereo information acquisition technology could providesupport for the environment detecting and intelligence survey strongly. And likewise,the technology is of great significance to estabilishing digital terrain model in thegeographic information system, and contributed greatly in areas such as nationaldefense and environment.Research of near space platform3-D information acquisition starts from thebasic theory of general three-dimensional information retrieval. And then, thebackground of the near space3-D information acquisition is discussed, from theperspective of the application environment. The basic theory of3-D informationacquisition of the intergral process is studied in this paper, based on the height ofnear space platform for information processing studies, to calculate the effect ofphotographing parameters to the accuracy of acquired information.The first step of3-D information acquisition is the image feature extraction andstereo matching of remote sensing images. Images collected by near space platformare usually with the diversification of the photographing angle, instable attitudes,large dynamic range of resolution etc. Algorithms of feature extraction and stereomatching for this kind of image, not only need to have a good scale invariantfeatures of algorithm, should also be good at extracting more important detailsinformation to feedback the platform information processing. Therefore, the SIFTfeature extraction method based on affine invariant model is studied in this paper,and the methods of fundamental matrix screening and HARRIS corner extraction are adopted for the joint optimized matching. And then the three-dimensional cornerdetection method based on graph grammar is presented to extract the key cornerscorresponding to the three dimensional key points in the real world.The three-dimensional coordinates solvedfrom the stereo matching results ofnear space remote sensing images, tend to be a point cloud model which is discreteand uneven. Hence, certain preprocessing should be taken to the point cloud data,and grid subdivision should be set up with corresponding model. This paperproposes a terrain mesh subdivision method based on a quadtree partitionapplication model algorithm, and the efficiency of the meshing is improved.There is rich texture information in remote sensing images, with which thetexture features analysis could aquire a lot of useful information. Combined with thegrid model, the extracted image texture from the images can provide intuitive visualdisplay, which would make it convenient for the man-machine interactive system tomake reasonable judgments. In this paper, a lot of texture features are extracted fromremote sensing data and semi-physical simulation data, and used in texture featuresanalysis based on gray level co-occurrence matrix and Gabor filter analysis. Theresults of the analysis are applied in the process of texture mapping, finally mappingthe texture orderly back to the original object target, to arhieve the3-D visualizationview.