A comparison of spacecraft collision detection strategies using a virtual conjunction dataset

Abstract

Conjunction assessment in the field of collision avoidance is customarily based on a single or a fusion of collision risk indexes. There are several risk indexes proposed in the literature, yet the authors could not find a comparative study of such indexes that was based on the miss detection performance. Miss detection is in fact one possible target parameter when choosing an actionable threshold and is a good candidate for high-traffic conditions. This work applies and compares four different detection strategies i) Probability of Collision, ii) Mahalanobis distance, iii) Wald Sequential Probability Ratio test and iv) Gaussian Belief Function on a common dataset of virtual conjunctions, measuring the performance on the Receiver-Operating-Characteristic (ROC) plane (detection rate – vs – false alarm rate). The virtual conjunction dataset offers an unprecedented realism in both the geometry and the true collision rate of candidate conjunctions given some orbital scenario. Using this dataset, we estimate the realistic performance of the detectors. Results indicate that all the examined risk indexes offer a performance linkable to the miss detection rate. However, the classical Probability of Collision can underperform with respect to the Mahalanobis distance detector, which might be preferred for its simplicity instead.