Faber, Weston Ryan (2018-05). Multiple Space Object Tracking Using A Randomized Hypothesis Generation Technique. Doctoral Dissertation. Thesis uri icon

abstract

  • In order to protect assets and operations in space, it is critical to collect and maintain accurate information regarding Resident Space Objects (RSOs). This collection of information is typically known as Space Situational Awareness (SSA). Ground-based and space-based sensors provide information regarding the RSOs in the form of observations or measurement returns. However, the distance between RSO and sensor can, at times, be tens of thousands of kilometers. This and other factors lead to noisy measurements that, in turn, cause one to be uncertain about which RSO a measurement belongs to. These ambiguities are known as data association ambiguities. Coupled with uncertainty in RSO state and the vast number of objects in space, data association ambiguities can cause the multiple space object-tracking problem to become computationally intractable. Tracking the RSO can be framed as a recursive Bayesian multiple object tracking problem with state space containing both continuous and discrete random variables. Using a Finite Set Statistics (FISST) approach one can derive the Random Finite Set (RFS) based Bayesian multiple object tracking recursions. These equations, known as the FISST multiple object tracking equations, are computationally intractable when solved in full. This computational intractability provokes the idea of the newly developed alternative hypothesis dependent derivation of the FISST equations. This alternative derivation allows for a Markov Chain Monte Carlo (MCMC) based randomized sampling technique, termed Randomized FISST (R-FISST). R-FISST is found to provide an accurate approximation of the full FISST recursions while keeping the problem tractable. There are many other benefits to this new derivation. For example, it can be used to connect and compare the classical tracking methods to the modern FISST based approaches. This connection clearly defines the relationships between different approaches and shows that they result in the same formulation for scenarios with a fixed number of objects and are very similar in cases with a varying number of objects. Findings also show that the R-FISST technique is compatible with many powerful optimization tools and can be scaled to solve problems such as collisional cascading.
  • In order to protect assets and operations in space, it is critical to collect and maintain accurate
    information regarding Resident Space Objects (RSOs). This collection of information is typically
    known as Space Situational Awareness (SSA). Ground-based and space-based sensors provide information
    regarding the RSOs in the form of observations or measurement returns. However, the
    distance between RSO and sensor can, at times, be tens of thousands of kilometers. This and other
    factors lead to noisy measurements that, in turn, cause one to be uncertain about which RSO a
    measurement belongs to. These ambiguities are known as data association ambiguities. Coupled
    with uncertainty in RSO state and the vast number of objects in space, data association ambiguities
    can cause the multiple space object-tracking problem to become computationally intractable.
    Tracking the RSO can be framed as a recursive Bayesian multiple object tracking problem with
    state space containing both continuous and discrete random variables. Using a Finite Set Statistics
    (FISST) approach one can derive the Random Finite Set (RFS) based Bayesian multiple object
    tracking recursions. These equations, known as the FISST multiple object tracking equations, are
    computationally intractable when solved in full. This computational intractability provokes the
    idea of the newly developed alternative hypothesis dependent derivation of the FISST equations.
    This alternative derivation allows for a Markov Chain Monte Carlo (MCMC) based randomized
    sampling technique, termed Randomized FISST (R-FISST). R-FISST is found to provide an accurate
    approximation of the full FISST recursions while keeping the problem tractable. There are
    many other benefits to this new derivation. For example, it can be used to connect and compare the
    classical tracking methods to the modern FISST based approaches. This connection clearly defines
    the relationships between different approaches and shows that they result in the same formulation
    for scenarios with a fixed number of objects and are very similar in cases with a varying number
    of objects. Findings also show that the R-FISST technique is compatible with many powerful
    optimization tools and can be scaled to solve problems such as collisional cascading.

publication date

  • May 2018