Information-Driven Search and Track of Novel Space Objects

TL;DR

This work tackles rapid acquisition of newly detected space objects by combining a CPHD multi-target tracker with information-based sensor control under Admissible Region constraints. By representing the search set as a CPHD intensity $D_k(\boldsymbol{x})$ with a cardinality distribution $\rho_k(n)$ and augmenting it with a clutter model anchored to catalog objects, the approach robustly handles false positives, missed detections, and negative information from empty scans. A novel GMM splitting method accounts for the discontinuity in detection probability at the FOV boundary, while a Rényi-divergence-based reward (via a particle CPHD representation) guides optimal follow-up sensor steering in a POMDP framework. Numerical case studies show that the information-driven strategy outperforms naive scanning in converging to the true target set and reducing uncertainty, demonstrating practical gains for space-domain awareness with optical sensors.

Abstract

Space surveillance depends on efficiently directing sensor resources to maintain custody of known catalog objects. However, it remains unclear how to best utilize these resources to rapidly search for and track newly detected space objects. Provided a novel measurement, a search set can be instantiated through admissible region constraints to inform follow-up observations. In lacking well-constrained bounds, this set rapidly spreads in the along-track direction, growing much larger than a follow-up sensor's finite field of view. Moreover, the number of novel objects may be uncertain, and follow-up observations are most commonly corrupted by false positives from known catalog objects and missed detections. In this work, we address these challenges through the introduction of a joint sensor control and multi-target tracking approach. The search set associated to a novel measurement is represented by a Cardinalized Probability Hypothesis Density (CPHD), which jointly tracks the state uncertainty associated to a set of objects and a probability mass function for the true target number. In follow-up sensor scans, the information contained in an empty measurement set, and returns from both novel objects and known catalog objects is succinctly captured through this paradigm. To maximize the utility of a follow-up sensor, we introduce an information-driven sensor control approach for steering the instrument. Our methods are tested on two relevant test cases and we provide a comparative analysis with current naive tasking strategies.

Figures (14)

• Figure 1: AR semimajor axis and eccentricity constraints for the GEO (left) and GTO (right) test cases described in Section \ref{['sec:results']}. The intersection of the constraint sets is the feasible ranges and range rates associated to the measurement.
• Figure 2: GMM approximation of the intersection of the constraint sets for the GEO (left) and GTO (right) test cases, described in the results section.
• Figure 3: Illustration for the recursive splitting operation used in this work. Here we show the original mixand and its resulting mixture after one and two splits.
• Figure 4: Here, we illustrate the plug-in kernel intensity estimate in a 1D example for the prior (left) and posterior, given an empty measurement (right). These are used to generate the Rényi-divergence estimate in Eq. \ref{['eqn:alpha_div_CPHD_particle_approx']}.
• Figure 5: Joint sensor control and multi-target filtering flowchart for seeking and tracking newly detected space objects.