A Novel Approach for Tracklet-Object Association Using a Pseudo-Probability Metric

Kurzfassung

Passive-optical observations of resident space objects result in tracklets with astrometric angle pair measurements. These tracklets are checked against a base catalogue for possible associations to already known objects. The association process may be ambiguous, as the measurements in tracklets carry uncertainties and/or satellites may have manoeuvred. This ambiguity can either occur in the form of an association of one tracklet to multiple objects or of multiple tracklets to one object. Furthermore, it is possible that no catalogue object can be identified for a given object, even if a tracklet was the result of a follow-up observation. All of these possibilities may lead to misleading associations or none at all, and thus the orbital information of the corresponding objects may not be updated. Commonly, the trackletobject association is performed by calculating the Mahalanobis distance between the measurements and the propagated positions of the objects. Within the sensor network SMARTnet (Small Aperture Robotic Telescope Network) and the attached database BACARDI (Backbone Catalogue of Relational Debris Information), a new approach is applied: each tracklet-object combination is assigned a number reflecting a likelihood value for this combination to be true. This value is derived by comparing the measured positions to propagated positions based on catalogue orbits. The deviation from a perfect match is derived via coordinates in a normalized vector space and the norm, i.e., the distance to the origin, is converted into the likelihood value, called pseudo-probability. The pseudo-probability values of all possible trackletobject associations for a given tracklet are stored. In the subsequent orbit determination process, the measurement residuals within each tracklet are assessed for all candidate tracklet-object combinations, and the combination resulting in the lowest residual RMS is selected. In case of all residuals being too large, the tracklet will be stored without any association in the database for further analysis. We will describe the method in detail and show the advantages compared to currently used methods. In addition, we will discuss some challenges that have to be overcome for a fully automated process.