Multi-Target Tracking for SMARTnet: Multi-Layer Probability Hypothesis Filter for Near-Earth Object Tracking

Kurzfassung

In this paper, a modified version of the finite set statistics-based Probability Hypothesis Density (PHD) filter is developed specifically for the optical multi-target tracking of objects in the near-Earth realm for Space Situational Awareness (SAA). A two-step PHD filter is proposed in a modified version. One labeled PHD filter is used on the orthogonal image plane, in which linear dynamics in a fourparameter state is employed, forming so-called tracklets. Tracklets are associated sets of a few closelyspaced observations covering a negligible part of the overall orbit. Furthermore, tracklets are fed into a second PHD filter in a modified measurement update version, utilizing the full near-Earth astrodynamics with a six parameter state. In the modification, each tracklet leads to only one update in the PHD, but all observations within the tracklet are processed in the single target Markov transition process within the filter. In this case, the single target filter is an Extended Kalman Filter. In addition, the birth process that has been usually in typical SSA applications shifted to the birth step, forcing a data-driven birth with the disadvantage of a severe model mismatch, back to the propagation step, as in the original PHD filter formulation, avoiding the mismatch. In order to overcome the lack of probabilistic description availability (one of the triggers of the shift to the datadriven update step of previous authors), the data is preprocessed. This has the advantage that birth can employ traditional initial orbit determination methods and does not have to rely on the initialization with an incomplete state using, e.g., an admissible regions approach. The results are generated using the optical data of the DLR SMARTnet telescope network and are compared to the DLR BACARDI data processing.