Multi-Object Filtering for Space Situational Awareness

摘要

The first phase of this project focused on the exploitation of the novel high-order regional statistics for Random Finite Set (RFS)-based multi- object filters, estimating the size of the target population, with associated uncertainty, and in any desired region of the surveillance scene. First phase concluded that: (1) Regional statistics are a promising tool for the assessment of situational awareness because they are able to estimate the performance of the multi-object filter in any desired region of the state space; (2) RFS-based multi-object filters, however, have limited use since they do not propagate individual information on targets, but only collective information on the target population (i.e. no tracks are maintained). The second phase explored the construction of a novel filtering solution for multiple-target tracking with the following core requirements in mind: The solution must be derived from a well-defined and well-identified probabilistic framework; It must be compatible with the higher-order regional statistics, and more generally with the available statistical tools for the assessment of RFS-based multi-object filters; It must maintain specific information on identified targets (i.e. tracks). A novel multi-object filtering framework, describing the objects of interest through the concept of stochastic population, has recently been proposed and allows the construction of a novel class of multi-object filters fulfilling the requirements above. This second phase illustrates this multi- object filtering framework on the tracking of multiple targets on simulated orbital scenarios driven from SSA problems. Unlike those derived from the FISST framework, the multi-target detection and tracking algorithms derived from this alternative framework maintain an inherent history of past estimates and past observations for each potential target identified through at least one detection across the scenario.