On The Radar Target Tracking Method Based On Compressed Sensing Theory

With the rapid development of the modern shipping business and frequent ship trade in countries and regions,the quantity of ship increased sharply,making water traffic situation severer and safe navigation issues increasingly prominent.In addition,higher demands are proposed for the VTS system,especially for accurate tracking of ship.In view of tracking instability,target glint and strong occlusion of the traditional radar tracking system,this thesis utilizes the high data processing ability of compressed sensing theory to apply the high dimension feature information of the target image to the conventional radar target tracking process.Make use of the advantage of visual tracking and Kalman filter technology to track the target.Thereby,to improve tracking performance.Firstly,this thesis analyzes and demonstrates in detail that the radar target image may be sparse from the perspectives of image histogram and correlation.Then,the basic characteristics of radar target image are analyzed and the author presents a relatively simple but effective described method with "Texture" feature.Moreover,in order to further improve the matching rate,this thesis fully considers radar target position,heading and other geometric features when it carries out "texture" feature matching at a later stage.Secondly,this thesis introduces the compressed sensing theory and chooses a more simple but efficient sparse base and measuring matrix,and demonstrated its excellent performance indirectly through reconstruction algorithm.At last,based on preliminary analysis and feasibility studies,the author proposes a representation method for target feature based on compressed sensing theory.By this method,sparse random measurement matrix is utilized to reduce the dimension for the samples collected with high-dimensional and multi-scale features.Not only the significant features of target image and the space structure are retained in the low-dimensional feature space,but also greatly reduces the feature storage space.What's more,in the target tracking process,the thesis applies uniform linear motion model and discrete Kalman filtering technique to predict the position of the target.Then,the image tracking technology is used in the predicted position,taking the target detection process as a binary classification task with local search.Regard the target position corresponding to the classifier maximum value as the measured value of Kalman filtering process,and then get accurate estimation of target position combining predicted value with measured value.The experiments confirmed that the tracking algorithm is able to track the target continuously,stably and accurately,and maintain a reliable and stable target tracking in the face of approaching target interference,strong occlusion and the like problems.It has strong robustness.In addition,the introduction of compressed sensing theory not only reduces the computational complexity,and improves the tracking accuracy to some extent.