Research On Key Techniques Of Stereo Matching For Stereo-Radargrammetry

Stereo-Radargrammetry is one of the main techniques for SAR large topographic survey.Stereo-Radargrammetry uses homologous points to measure ground threedimensional information between the two images acquired from different incidence angles.However,SAR image has inherent speckle noise and there are large geometric difference and radiation difference between stereo SAR images,which leads to the difficulty of extraction of the homologous points between SAR stereo images.Therefore,Study on matching algorithms for SAR stereo images and automatically get homologous points,which has significanes to improve the accuracy of StereoRadargrammetry and to promote the automation of Stereo-Radargrammetry for large topographic survey.Based on the above,the dissertation carry out three aspects of research in stereo SAR images epipolarity and multi-feature matching and correlation-based matching algorithms considering information of matching window.The main work and innovations of the dissertation are as follows:(1)SAR epipolar line equation are firstly theoretical derived based on the definition of epipolar line on the remote sensing image and SAR Range-Doppler model.We study properties of approximate epipolarity in spaceborne SAR images,and an epipolar rectification algorithm for satellite SAR images is proposed,in the processing of which the transformation between slant range and ground range satellite SAR images is introduced.We also study properties of approximate epipolarity in airborne SAR images and the relationship between slant range and ground range airborne SAR images,and an epipolar rectification algorithm for airborne SAR images is also proposed on the basis.The experiment results demonstrate that the algorithms can eliminate vertical parallaxes.An epipolar model based on projection reference plane is introduced,and a modified epipolar rectification algorithm for satellite SAR images is developed based on this model.It introduce an external DEM in the process based on epipolar lines are approximate parallel in different projection reference planes.The experiment result show that it can eliminate vertical parallaxes and lower the effect of geometric distortion of SAR images and reduce the search area at horizontal parallaxes.(2)The accuracy of gradient orientation of each feature and the distinguishing ability of each feature descriptor are affected by inherent multiplicative speckle noise in SAR images.The feature matching is susceptible to the similar textures,and the number of matching points obtained by single feature matching algorithm are less and its matching success rate is low.Based on it,multi-feature matching algorithm based on triangle multi-conditional constraints is proposed.This algorithm use maximally stable extremal regions(MSER)algorithm and scale invariant feature transform(SIFT)algorithm to obtain more stable feature matching points based on the ratio gradient instead of difference gradient,then the remaining features are matched by triangular area and triangular scale and triangular direction constraint conditions,which adopted by using stable matching points.The algorithm was applied on the airborne stereo SAR images and the result demonstrate that it can effectively increase right matching points and improve matching rate.Considering there are many noise points in the features caused by the low signal to noise ration of SAR image,and matching features are located on rich region of texture information in SAR images,an algorithm for stereo SAR images based on geometric constraint is proposed.SIFT matching algorithm was applied on the multi-look images and homography geometry constraint with the grid controlled were utilized to get more and well-distributed matches.The experiment demonstrate that the proposed algorithm has significant improvement in terms of the amount,precision and distribution of the matches.(3)Considering the influence of geometric distortion in stereo SAR images,a matching algorithm with adaptive window based on epipolar constraint is proposed.Image parameters and an external DEM of coarse resolution are utilized to reduce the search area alone the epipolar line.Then do Summed Normalized Cross Correlation(SNCC)matching,in the procedure of which the matching window is adaptive extended to lower the effect of geometric distortion in stereo SAR images.Experiment on the TerraSAR-X spotlight stereo images,the result show that the proposed algorithm can increase the account of correct homologous points in case of mountainous areas and high contrast area and improve the correct matching rate.A matching algorithm based on disparity maps fusion is proposed.Considering the effect of texture information,various disparity are computed using several different matching window sizes based on Normalized Cross Correlation(NCC),To evaluate their confidence level,we propose a new confidence measure using the advantage of left right consistency and signal to noise.Then a newly proposed disparity maps fusion strategy is used for formation of disparity maps computed by several different matching window sizes.This disparity maps fusion strategy considers not only the confidence level of the disparity itself but also its neighbors.The results demonstrate that the accuracy of DEM generated with the proposed algorithm is improved.(4)Ascending stereo images and descending stereo images that coverd same area are used to generate DEM.An algorithm of tie points automatically acquired for multi-strips images in regional network is proposed based on the research of multifeatures matching,and experiment show that it can provide sufficient tie points for block adjustment.Then an automatic image matching algorithm is proposed based on the research of SAR epipolar images and SAR correlate matching,and the accuracy of DEM derived from the method is about 7 meters.