Research On Registration Of WAS SAR-GMTI Radar Large View Image Sequences

In the wide-area surveillance, the SAR-GMTI system is used to acquire the staticinformation of the wide and interested region, and to monitor the goal of sports day,all-weather. Forming moving targets’ trajectory by the wide-area surveillance mode, itfaces the spatial mismatch problems, such as resolution inconformity and geometrydistortion, between images of the same scene obtained by different views. And theSAR image contains lots of speckles, which makes it difficult to match the SAR imagesequences. In order to realize the registration of large view image sequences, thisdissertation mainly deals with the research on the algorithm for SAR image positioningand images matching to achieve the registration of the wide-area surveillance SARimage sequences. The main contents include:(1) The geometric model of SAR imaging is presented. According to range andDoppler equations, the positioning model of SAR image is deduced, which could beused for the registration of SAR images by different views.(2) Two algorithms of image registration based on the knowledge of the graylevel and based on the characteristics are studied. It introduces several classical imageregistration algorithms based on the gray level, and the registration algorithm based onscale invariant feature transform is mainly introduced.(3) A registration algorithm combined prior information of SAR images ispresented, which could be used to matching large view SAR image sequence ofwide-area surveillance. Firstly, it detects edge corners by using edge corner detectionalgorithm. Secondly, the location of corners is conducted with SAR imagingparameters and platform navigation positioning system parameters. Also, it is used tohelp to determine candidate matching corner area of in registration. Finally, it starts tosearch matching corners in the candidate area by the least Euclidean distance. Theexperiments compared to SIFT algorithm verify the performances in the stability,efficiency and registration accuracy of the proposed algorithm.