Shahzad Virani, Daniel Guggenheim School of Aerospace Engineering – Georgia Institute of Technology, Timothy S. Murphy, Daniel Guggenheim School of Aerospace Engineering – Georgia Institute of Technology, Marcus J. Holzinger, Daniel Guggenheim School of Aerospace Engineering – Georgia Institute of Technology, Brandon A. Jones, Department of Aerospace Engineering and Engineering Mechanics – University of Texas at Austin
Keywords: SSA, Detectionless Tracking, Data-Driven, FiSSt, Multi-Bernoulli, Real-time
Abstract:
Autonomous search and recovery of resident space object (RSO) tracks is crucial for decision makers in SSA. This paper leverages dynamic data driven approaches to improve methodologies used in real-time detection and tracking of RSOs with a low signal-to-noise ratio (SNR). Detected RSOs are assigned to be tracked using one of two simultaneously operating algorithms. The Gaussian Mixture Proability Hypothesis Density (GM-PHD) filter tracks all RSOs above a certain SNR threshold, while a Detectionless Multi-Bernoulli filter (D-MB) detects and tracks low SNR objects. The D-MB filter uses matched filtering for likelihood computation which is highly non-Gaussian for dim objects. Hence, the D-MB filter is particle based which leads to higher computational complexity. The primary idea proposed in this paper is to balance the computational efficiency of GM-PHD and high sensitivity of the D-MB likelihood computation by dynamically switching tracks between the two filters based on the SNR of the target; allowing for real-time detection and tracking. These algorithms are implemented and tested on real data of objects in the geostationary (GEO) belt using a wide field-of-view camera (18.2 degrees). A star tracking mount is used to inertially stare at the GEO belt and data are collected for 2 hours corresponding to RSOs being observed in 48.2 degrees of the GEO belt.
Date of Conference: September 19-22, 2017
Track: Space Situational Awareness