Research On Video SAR Moving Target Detection And Tracking

In recent years,with the development of high-frequency devices and the improvement of signal processing ability,video synthetic aperture radar(SAR)has made great progress.As a new radar imaging and detection technology,video SAR makes radar imaging have the ability of dynamic information indication close to optical and infrared technology for the first time.Differing from traditional microwave SAR,video SAR usually works in extremely high frequency(EHF)and above.Its imaging characteristics of high resolution,high frame rate and low delay are conducive to the real-time detection of ground moving targets.The video SAR system is portable and flexible,and its technical characteristics of organic combination of imaging and moving target detection are particularly conducive to the equipment of the observation and combat integrated UAV that is vigorously developed at present.Therefore,as the core technology of video SAR,moving target detection has attracted extensive attention.Based on the topics of video SAR moving target monitoring,this paper carries out the research on moving target detection and tracking.Combined with practical problems,a series of moving target detection and tracking methods for video SAR are proposed.The proposed methods are analyzed and discussed by means of theoretical calculation and simulated/measured data verification.The main research contents are as follows:1.Combined with the imaging characteristics of video SAR,the imaging characteristics of moving target energy and shadow are analyzed and discussed quantitatively,which provides a theoretical basis for detection and tracking.The analysis mainly includes the relationship of video SAR imaging parameters,the position shift and defocus characteristics of moving target energy,the formation mechanism and parameter dependence of moving target shadow,shadow detection performance,and so on.2.For moving target shadow,two detection algorithms and a track-before-detect algorithm are proposed.The detection algorithm based on regional evaluation effectively alleviates the influence of noise and improves the detection performance of moving target shadow through shadow confidence scoring and double threshold processing.The detection algorithm based on background subtraction extracts the dynamic shadow through image processing methods such as inter frame registration,background modeling and subtraction,which effectively reduces the influence of complex scenes and noise.The track-before-detect algorithm based on super-pixel segmentation not only alleviates the computational explosion but also effectively improves the overall tracking performance of moving target shadow by using the intensity and shape characteristics of super-pixel blocks.3.To improve the robustness of moving target tracking,a tracking framework based on joint track association is proposed.It mainly includes pre detection processing based on the designed detectors,joint track association algorithm,and tracking solutions for typical situations of maneuvering targets.The proposed framework can effectively reduce the false alarm rate and missed detection rate by making full use of the space-time information of moving target shadow and energy.Aiming at the typical situations of moving targets in reality,a series of track reconstruction and update schemes are proposed.The verification experiments based on simulated/measured data show that the proposed framework can effectively decrease the false alarms and missed detections caused by target mobility in the single domain methods.The robustness of video SAR moving target tracking can also be improved.4.To make full use of the features of moving targets,solve the existing tracking drift/collapse problems,and improve the real-time tracking performance,an online video SAR moving target tracking framework based on a joint kernel correlation filter is proposed.It mainly includes joint domain target initialization,feature extraction and training by dual kernel correlation filter,interactive tracking correction processing,feedback processing,and adaptive update strategy.Through fast kernel feature training and interactive correction and update processing between domains,the problems of tracking drift and collapse are effectively solved,and the tracking success rate and accuracy are significantly improved.The performance improvement of the proposed framework is verified by measured data and comparative experiments.