A particle-based search strategy for improved Space Situational Awareness

Abstract

In certain tracking applications, it is not sufficient to assume that the measurement of a target's state can be made whenever a sensor is tasked to do so. For example, the target's position may lie outside the sensor's limited field of view. Nevertheless, failure of this sort still yields some information. It tells us where the target is not. This information is difficult to capture in conventional filtering. In the context of catalogue maintenance of resident space objects, a central task in Space Situational Awareness, we demonstrate how the particle filter may be adapted to account for occasional failed observations and to guide the process of target reacquisition while maintaining a high quality track at other times. The results of a numerical simulation show that while an Unscented Kalman Filter can lose track of objects in more challenging circumstances, the proposed particle method consistently reacquires and tracks all objects.

Analysis of Information-Theoretic Initial Sensor Search Method for Space Situation Awareness

“…The most straightforward approach to this initial search problem is a maximum-probability approach [11,15,16]. This approach points the sensor at the most probable candidate direction to capture the object in the FOV.…”

“…For analysis of the maximum-probability approach and the information-theoretic approach, we consider the same LEO-based single-sensor single-object search problem studied in the literature [17]. Following the convention in the past literature [11,[15][16][17][18], a binary sensor model is used where the sensor provides a binary detection measurement, i.e., either a detection or no detection measurement related to the presence of the object in the FOV; known nonzero probabilities for false alarms and missed detections are assigned. This simple model is expected to reflect the context of the initial sensor search, where our goal is to obtain observations of the object as soon as possible and as accurately as possible.…”

“…This simple model is expected to reflect the context of the initial sensor search, where our goal is to obtain observations of the object as soon as possible and as accurately as possible. In addition, we also follow the literature [11,[15][16][17][18] and use a particle representation for the object's state probability distribution; these particles provide dual roles in approximating the distribution as well as discretizing the control search space defining how to steer the sensor at each observational time step.…”