Steven Gehly (University of Colorado at Boulder), Dr. Brandon Jones (University of Colorado at Boulder), Dr. Penina Axelrad (University of Colorado at Boulder)
Keywords: AEGIS, CPHD, SSA, probability of detection
Abstract:
The problem of space situational awareness (SSA) involves characterizing space objects subject to nonlinear dynamics and sparse measurements. Space objects in GEO are primarily tracked using optical sensors, which have limited fields of view, imperfect ability to detect objects, and are limited to taking measurements at night, all of which result in large gaps between measurements. In addition, the nonlinear dynamics result in state uncertainty representations which are generally non-Gaussian. When estimating the states of a catalog of space objects, these issues must be resolved within the framework of a multitarget filter. To address the issue of non-Gaussian uncertainty, the Adaptive Entropy-based Gaussian-mixture Information Synthesis (AEGIS) filter can be used. AEGIS is an implementation of the Unscented Kalman Filter (UKF) using an adaptive number of Gaussian mixture components to approximate the non-Gaussian state probability density function (pdf). Mixture components are split when nonlinearity is detected during propagation, typically during long data gaps, and can be merged or removed following measurement updates to reduce computational effort. Previous research has examined the use of AEGIS in multitarget filters based in Finite Set Statistics (FISST), including the Probability Hypothesis Density (PHD) filter and Cardinalized PHD (CPHD) filter. This paper uses the CPHD filter because in other applications it has been demonstrated to be more effective at estimating and maintaining the cardinality, or number of objects present, when objects are often leaving the sensor field of view (FOV). An important consideration in implementing the filter is the computation of the probability of detection. Existing formulations use a state-dependent probability of detection to assign a value based on whether the mean estimated state is in the sensor FOV. This paper employs a more realistic development by mapping the full state pdf into measurement space and integrating over the area within the sensor FOV. Results are provided for a simulation using the AEGIS-CPHD filter to track multiple objects at GEO, subject to sparse measurements and including clutter, in order to demonstrate the filters ability to maintain custody of space objects in a realistic SSA scenario. A quantitative comparison is included using the filter with both the state-dependent and state pdf determined probability of detection. References:
Date of Conference: September 9-12, 2014
Track: Astrodynamics