Maneuver Detection with Two Mixture-Based Metrics for Radar Track Data

In the context of space situational awareness, orbit determination and maneuver detection pose considerable challenges given the ever increasing number of noncooperative satellites that perform unspecified maneuvers at unknown epochs. A new metric is proposed for maneuver detection here, based on measurements from a single surveillance radar. Data from the radar track are leveraged to derive this metric; it can flag the occurrence of a maneuver since the last time a satellite was observed. In the scenarios considered in this work, the main challenge is having just a few available observations in between long propagations. Thus, a realistic quantification of uncertainty becomes crucial; Gaussian mixtures are chosen for this work. The propagated state is compared against the radar track data to compute a well-known cost metric, the evidence in Bayes's update, similar to the Mahalanobis distance (MD). The contribution of this work lies in using the variation of the cost metric across track points to define a new maneuver detection metric. The focus is on testing this new metric in diverse and realistic scenarios to gain practical intuition about its performance and limitations. The preliminary results are promising, simultaneously improving upon the classical MD at maneuver detection and reducing false positives.