Research On Target Tracking Technologies For Space-based Optical Surveillance System

The space has become the strategic frontier for maintaining the national security and interest. The capability of target acquisition and tracking is the foundation of the space dominance and controlling. Space-based optical surveillance system combines the advantages of space-based observer and optical sensor, and has the superiority in many aspects, such as unlimited by country boundaries, wide aera survelliance, high precision measurement, self-hiding, and so on. Consequently, space-based optical surveillance system has been studied comprehensively by many acadamicians. This dissertation focuses on the target tracking problem based on such surveillance system. The key technologies are discussed and researched particularly, including single target tracking on focal plane, multiple target tracking on focal plane, track-to-track association under multiple observers and multiple sensors, target tracking in the space. The main contributions of this dissertation are demonstrated as follows:In chapter 2, the target kinematic characteristics are analyzed firstly, and then the main aspects of the target dynamic model are discussed. Secondly, definition and transformation of basic coordinates are introduced; a measurement model is established based on the mapping relationship from target stereo position to focal plane. Thirdly, an analytical method of the line-of-sight (LOS) measurement error is proposed for space-based optical sensor, a LOS error model is given. Lastly, theoretic target location accuracy is derived under the model. The aforementioned researches can serve as the foundation for the studies of succeeding chapters.In chapter 3, the issue of single target tracking on focal plane is studied. Firstly, the measurement model and target dynamic model on focal plane are both established. Secondly, a single target tracking algorithm on focal plane is proposed under the random finite set (RFS) theory framework, which avoids data association based on modeling target states and measurements as RFS variables. Thirdly, the algorithm implementation problem is taken into account; a Sequential Monte Carlo (SMC) implementation and a Gaussian Mixture (GM) implementation is presented, respectively. Simulation results demonstrate that the tracking performance of this algorithm is superior to the traditional single target tracking algorithm IPDA.In chapter 4, the issue of multiple target tracking on focal plane is studied. The arisen probability hypothesis density (PHD) filter in recent years is applied to meet this problem, and some approaches are studied to improve the PHD filter performance on focal plane from the aspects of tracking accuracy, computation efficiency, and PHD implementation. Firstly, by using the signal amplitude information (AI), an AI auxiliary PHD filter on focal plane is proposed, which enhances the target tracking accuracy in low signal-to-noise ratio (SNR) circumstance. Secondly, by introducing the gating technology within traditional multiple target tracking methods, a gated PHD filter on focal plane is proposed, which improves the computation efficiency in strong clutter circumstance. Finally, by integrating the sigma-point filter and Gaussian mixture models (GMM) into PHD, a GM sigma-point PHD filter is proposed, which extends the GM-PHD from linear Gaussian conditions to nonlinear nonGaussian conditions.In chapter 5, the issue of track-to-track association under multiple observers and multiple sensors is studied. Firstly, the global nearest neighbor (GNN) principle used in measurement-to-measurement association problem is introduced into the hinge angle difference hypothesis testing method, a track-to-track association algorithm based on 2 dimensional assignment principle is proposed, treating the hinge angle difference as the cost. Simulation results demonstrate that the performance is better than the hypothesis testing method. However, the hinge angle difference based track-to-track association algorithms mentioned above have some drawbacks, such as the performance is influenced by the geometry easily, and may become invalid at certain conditions. Because the target kinetic characteristics is a valuable information, hence a maximum likelihood (ML) based track-to-track association algorithm is proposed, which makes the best of the target kinetic information. At the same time, the likelihood ratio presentation and target state ML estimation problems are also studied. Simulation results show that the performance of the ML based track-to-track association algorithm is better than the hinge angle difference based track-to-track association algorithm, at the cost of additional computation.In chapter 6, the issue of target tracking in the space is studied. Firstly, a booster kinetic model is established, which considers the thrust acceleration variation and the attack angle variation at the same time. This booster kinetic model extends the application scope of the constant attack angle kinetic model. Then, the multiple model concept and UKF filter are integrated into this tracking problem, a target tracking algorithm based on multiple model and UKF filter is proposed, which uses 2 kinetic models, one is the booster kinetic model, and the other is the J2 Keplerian model. Simulation results show that this algorithm can adapt to interstage separation and burnout time automatically, and gets better tracking performance than single model UKF filter and single model EKF filter.