Research On Small Baseline Stereo Matching With High Precision

The image matching problem of traditional photogrammetry in the complex terrain can besolved effectively by small baseline photogrammetry, but which need highly accurate parallax.So, the high-precision image matching is crucial to high precision positioning for small baselineimages. Based on the small baseline theory analyzed in detail, the connections betweenbase-height ratio, matching precision and ground positioning precision is discussed. Thetechniques of original image matching, high-precision image matching for small baseline imagesis researched deeply. The main achievements and innovations are summarized as following:1. The significance and background of the research on small baseline photogrammetryespecially high precision image matching is described, the related works at home and board areintroduced.2. The influence of base-height ratio and noise on ground positioning precision isresearched mathematically to use the usual correlation coefficient method. The experimentsshow that the loss of ground positioning precision can be compensated by matching precision forsmall baseline images.3. Based on the small baseline photogrammetry researched deeply, the connections betweenbase-height ratio,matching precision and ground positioning precision is discussed with the realimages and secondary data.The related experiments are designed to demonstrated that thecredible height information can be gained by improving the matching accuracy4. Based on the advances and drawbacks of the local matching methods and the globalmatching methods, the semi-global matching method is imported which has the better results inthe repeated pattern areas,shadow areas and object edge as small base-height ratio image prematching method. The experiments show that the semi-global method can gain the credibilityand stability in different matching areas.5. Based on the phase correlation elementary theory, the two frequency-domainhigh-precision image matching methods are discussed deeply with the local up-samplingtechnique and the SVD technique imported to the cross power spectrum gained by phasecorrelation. The experiments show that the high-precision image matching algorithms the paperpresented can gain high matching precision under certain conditions.