Moving Object Detection Method For VideoSAR With Background Deformation Between Frames

With the increasing demand for modern military and civilian air-to-ground moving target detection,VideoSynthetic Aperture Radar(VideoSAR)has become an important research direction in the field of SAR moving targets detection.Meanwhile,since the moving target shadows can accurately reflect the real position of moving targets,VideoSAR moving target detection technology based on shadows has attracted wide attention in academia.However,due to the movement of radar platform and the change of azimuth angle,VideoSAR images have background deformation between frames,as such the false alarm rate is high and the accuracy is low.Aiming at the shortcomings of the existing VideoSAR moving targets shadow detection methods,this thesis starts from two aspects: correcting inter frame geometric deformation and suppressing interference caused by inter frame scattering intensity changes.By reducing the impact of interframe background deformation on shadow detection of VideoSAR moving target,the detection accuracy is improved.Main research contents are as follows:1.The basic principle of VideoSAR is studied.Firstly,the imaging principle of VideoSAR is introduced,and its high frame rate characteristics is analyzed.Then,the basic characteristics of VideoSAR moving target is explained,including two parts: the echo and the shadow characteristics of moving target.Finally,the thesis introduces the definition of inter-frame background deformation,as well as the causes of interframe geometric deformation and interframe target scattering intensity variation.2.The problem of VideoSAR image registration with interframe background deformation is studied.Aiming at the problem of interframe geometric deformation and SAR images susceptible to noise,this thesis proposes a VideoSAR image registration method based on nonlinear diffusion and Shi-Tomasi.It leverages the nonlinear diffusion filtering to construct the scale space,as such the image details and edges in the scale space can be preserved effectively.Meanwhile,in the procedure of feature point detection,ShiTomasi corner detection algorithm based on first-order difference is utilized,and the noise robustness of image registration method is improved.Finally,the qualitative and quantitative indicators such as the matching map of feature point pairs,the chessboard fusion map,the matching number of feature point pairs and the registration error RMSE in the experimental results of two-frame and multi-frame images verify the effectiveness and advantages of the proposed method.3.The problem of VideoSAR moving target shadow detection with interframe background deformation is studied.In order to solve the problem of high false alarm rate and low accuracy in moving target shadow detection caused by interframe scattering intensity,the thesis proposed a shadow detection method based on Total Variation and Robust Principal Component Analysis.In this algorithm,the registered image sequence frames are utilized to construct the data matrix.By leveraging the characteristics of the background low rank in data matrix,the sparse and continuous in moving targets shadow,as well as the sparse and discontinuous in interference,the problem of shadow detection can be converted to the problem of Total Variation and Robust Principal Component Analysis.As a result,the interference in shadow detection is suppressed,the false alarm rate is decreased,and the detection accuracy is improved.Finally,the moving target shadow detection results of simulation and experimental results,as well as qualitative and quantitative indicators such as false alarm and F-measure,verify that the proposed method can effectively suppress interference,reduce false alarm and improve detection accuracy.