A Higher Dimensional Tensor Decomposition Framework for Data Association in LEO Tracking

摘要

Common track maintenance algorithms that employ Bayesian techniques to track an object suffer from exponential growth with an increase in the number of objects. This is exacerbated in environments with dense targets and low signal-to-noise ratio (SNR). This paper improves the Dynamic Joint Probabilistic Data Association (DJPDA) filter built upon the pseudo-Bayesian sub-optimal Joint Probabilistic Data Association (JPDA) filter and the Dynamic Tensor Analysis (DTA) incremental tensor decomposition. DJPDA reduces the so-called "curse of dimensionality" seen in the growing number of joint association events, or feasible events, in JPDA by utilizing DTA to replace the complete set of measurements with a low-dimensional summary, known as the core. However, in cases where there is a strict limit on the time required to complete association, the accuracy of DJDPA is limited by the tensor rank of the decomposition. This paper explores the Windowed JPDA (WJPDA) to circumvent this problem by associating a batch of scans instead of one, i.e., DTA is performed on multiple scans at once. This provides the possibility of increasing the rank of the tensor decomposition to increase the accuracy of the association.