Brandon Jones, The University of Texas at Austin, Noble Hatten, The University of Texas at Austin, Nicholas Ravago, The University of Texas at Austin, Ryan Russell, The University of Texas at Austin
Keywords: astrodynamics, multi-target filtering, state estimation, orbit propagation, numerical methods
Abstract:
This paper presents a multi-target tracker for space objects near geosynchronous orbit using the Gaussian Mixture Cardinalized Probability Hypothesis Density (CPHD) filter. Given limited sensor coverage and more than 1,000 objects near geosynchronous orbit, long times between measurement updates for a single object can yield propagated uncertainties sufficiently large to create ambiguities in observation-to-track association. Recent research considers various methods for tracking space objects via Bayesian multi-target filters, with the CPHD being one such example. The implementation of the CPHD filter presented in this paper includes models consistent with the space-object tracking problem to form a new space-object tracker. This tracker combines parallelization with efficient models and integrators to reduce the run time of Gaussian-component propagation. To allow for instantiating new objects, the proposed filter uses a variation of the probabilistic admissible region that adheres to assumptions in the derivation of the CPHD filter. Finally, to reduce computation time while mitigating the so-called “spooky action at a distance†phenomenon in the CPHD filter, we propose splitting the multi-target state into distinct, non-interacting populations based on the sensor’s field of view. In a scenario with 700 near-geosynchronous objects observed via three ground stations, the tracker maintains custody of initially known objects and instantiates tracks for newly detected ones. The mean filter estimation after a 48 hour observation campaign is comparable to the measurement error statistics.
Date of Conference: September 20-23, 2016
Track: Astrodynamics