Tracking Techniques For Midcourse Target Complex Via Space-based Infrared Sensors

The unity of space-based infrared sensors plays an important role in the National Missile Defense Systems for its uncompareble advantages, such as 24-hour service, multispectral detectors, wide-area and long-range surveillance. As its future development trend, the prospects of detecting and tracking ballistic missiles during their whole lifecycle via sensors on LEO satellites are unpredictable. Apparently, the kernel mission of such a system is to precisely track missiles while midcourse stage in order to guide antimissile weapons.Midcourse target compelx released from post-boost vehicles is a great obstacle for the signal processing system onboard. This dissertation dedicates to the techniques on tracking such target group with high density and large number.The major contributions are declared as follows:Chapter 2 investigates the features of midcourse target group on kinematics and infrared physics at the beginning, along with the characteristics of the space environment appeared as backgrounds on focal plane array of infrared sensor. With the Satellite Tool Kit(STK) and proper simplified imaging processes according to the real scene, a medium-fidelity system-generated tracking scenario is established. Under this scenario, the kinematic characteristics of target group on FPA and target-generated measurement features are analyzed throughoutly. Then the tracking difficulties aroused by closely spaced objects and pixel clusters are introduced and a framework for tracking processes is proposed. Moreover, the scenario established in this chapter provides simulation data for the coming chapters.Chapter 3 mainly studies on some filter algorithms which are capable of jointly estimating the kinematic states of CSOs centers and the extensions of CSOs under both clutter and low detection probability circumstance. Thus, the target group with uneven density and continuous splitting and merging can be partitioned into several sub-unities and be steadily tracked.At first, the state-of-art tracking algorithms for multiple targets and group targets are reviewed throughoutly. Under midcourse target group scenario, the shortcomings of those algorithms which are based on point-target hypothesis, including formation tracking, are analyzed. By finding the relationship between the extended targets and centroid group tracking, the filters for midcourse target group can be effectively established according to the Probability Hypothesis Density filter for extended targets. Then, a partition method for infrared measurements is proposed using pixel magnitudes additionally not only the distance between pixels. The major advantage for the filters with this partition is that the amount of target complex can be approximately estimated as early as possible, even when the targets are very close to each other.Along with the partition, two filter algorithms are studied successively which focus on estimating the kinematic states and shapes of CSOs jointly no matter whether the group splites or merges. The former one, called the RHM-GM-PHD filter, augments the state vector with extension parameters which is described by an elliptical Random Hypersurface Model, breaking the limitation of shape estimation of the GM-ET-PHD filter. As to the latter one, called the MM-GIW-PHD filter, the shape model used in GIW-PHD filter is modified firstly and then multiple modified GIW-PHD filters with different shape models are combined together resulting in a good performance on shape-varying estimation. The efficiencies and capabilities of the two filters are demonstrated in several simulated scenes including different clutter rates, relative velocities between targets as well as the nonlinear degrees of trajectories on FPA. At the end of this chapter, the performances of shape estimation based on elliptical RHM and random matrices are compared explicitly and the range of application with each shape model is declared.For further considerations, Chapter 4 works on precisely tracking individuals by the super-resolution technique and the estimation-to-track association method.For the state-of-art, the basic principles and performances of super-resolution techniques are reviewed firstly. After analyzing the feasibility of the sparse reconstruction theory for the CSOs resolution issue, a sparse representation of the CSOs imaging model is given by sampling pixel grids and a heuristic method is proposed to handle over-estimation of the sparsty aroused by violating intercolumn correlation. Simulation results demonstrate that the proposed algorithm works fine even when the signal-to-noise ratio is low, and is capable of resolving multiple targets. Moreover, it consumes less computation load and is easy to apply. Besides, a criterion for judging when to do the super-resolution procedure is given.At last, an estimation-to-track association algorithm based on iterated multiassignment procedure along with track management is proposed. This algorithm combines both the advantages of hard and soft assignment criteria, which is capable of tracking target group nomatter dense or sparse. The one-to-one assignment is adopted to initiate tracks as quickly and many as possible especially when target group is very dense. After the super-resolution, those unresolved and redundant estimations may be assigned several times to establish one-to-many and many-to-one associations. In order to avoid unproper associations, a competition criterion is proposed according to the shape estimation of CSOs. After several steps of validation, target splitting and redundant tracks are confirmed. Several simulations demonstrate the performances of the proposed association algorithm, as well as the performances of the whole tracking system. It is proven that by combining the super-resolution technique, the tracking system can reach higher precision, and achieve steady transitions from group tracking to individual tracking.