Research On Moving Target Detection In Star Images

With the development of human activities in space,space debris has been increasing.The probability of collision between orbiting spacecrafts and space debris has also increased,and normal operations of spacecrafts have been seriously threatened.Therefore,the observation and positioning of space targets have important research value.In this thesis,by means of theoretical derivation,algorithm simulation,and hardware verification,the star-moving target detection and tracking system based on visible optical images is studied,and key technologies of target detection system such as registration of images,detection,tracking and location of the targets are completed,and The detection of moving objects in stars map is realized.The main research content consists of the following points:1.A fast registration algorithm based on triangle matching is proposed.Firstly,the background is suppressed by filtering and reference stars are extracted.Then,reference triangles are established among reference stars.The mapping relationship are determined by the triangles matching results of two consecutive frames.Finally,registration parameters are estimated via least squares method.Experimental result indicates that the position accuracy of the proposed triangle-based registration algorithm is 0.19 pixel RMSE and the average processing time of each image is around 300 ms on embedded system,which is twice faster than the current implementation.2.A target detection algorithm based on trajectory similarity analysis is proposed.Firstly,a motion model of space debris in star images is established.A target preselection mechanism based on multi-frame projection is implemented.The target linkage graph,with each of its node and edge representing a preselected target and a possible trajectory respectively,is established based on motion restrictions.The trajectory correlation matrix is derived.The trajectory correlation graph is established based on binarized trajectory correlation matrix by an adaptive thresholding algorithm.The trajectories are clustered by finding connected subgraph of the trajectory correlation graph.Unclustered trajectories are removed from target linkage graph.The remaining nodes in the linkage graph are expected to be targets.Abundant experiments have proved its superiority to a state of art method.The detection rate of 1371 targets in 240 consecutive images is 0.98,with the proposed detection method.Compared with current implementation,the proposed method improve the detection rate by 2%.3.Finally,the detection system is implemented on hardware platform consisting of FPGA and DSP.Experimental results indicates that this system is capable of real-time image processing with the maximum framerates of 1fps.After simulation and platform verification,the space target detection and tracking algorithm based on improved frame sequence projection and linear trajectory detection presented in this thesis is stable.It satisfies the design requirements for detection of weak targets with an SNR of 5 dB greater than 90% and meets the needs of actual project with an error less than 0.5 pixels.