Track-before-detect (TBD) is a statistical hypothesis testing technique that tests for the presence or otherwise, of a dynamic target using noisy and/or incomplete sensor measurements. Rather than making a detection decision directly from the observations, then handing relevant detection information to a model based tracker, TBD methods reverse this order of processing by assuming the presence of a target, and use information from the tracker to determine whether a target is present or not. Classically, tracking and TBD algorithms apply the Markov assumption to the target dynamics. At finer time scales the assumption is generally appropriate, and many well-known Markov based tracking and TBD algorithms exist. The work presented is motivated by applications with coarser time scales where the Markov assumption is less valid generally because real-world targets have a defined origin and destination. We will call such targets source-destination aware, and model them using Hidden Reciprocal Chains (HRCs). We examine whether the use of dynamic target models such as HRCs can improve TBD performance.
George Stamatescu received his B.Eng (EEE) and B.Ma&C.Sci from the University of Adelaide in 2013. Since 2014 he is a PhD candidate with the School of EEE, with Prof. Lang White. He has been supported by the D2D CRC since 2015. His research interests include statistical signal processing, target tracking and artificial intelligence.