Stochastic flows: the foundation of multi-object filtering with point processes

Though largely unknown to the multi-target tracking community, some of the most significant developments in the application of point process theory to multi-object filtering are found in the literature published in the late Soviet Union, which culminated with a publication of Detection of Moving Objects in 1980. This major work is regularly brought up in relevant Russian-language publications, yet it is rarely contrasted to the works on multi-object filtering produced in the English-speaking world. In turn, many of the original results that led to it were translated into English at the time, but subsequently have been rarely, if ever, cited in the English language literature. In fact, very few researchers are aware of those translations, including, until recently, the authors of this article.

Highly advanced for their time, these Soviet developments do not now necessarily promise a breakthrough in multi-object filtering as a number of central results have since appeared in the western literature without reference to the originators of these works. For example, Achkasov's 1971 multi-target likelihood for point processes, and Bakut and Ivanchuk's 1976 method for propagation of the intensity function of a point process prior have both appeared. Nevertheless, a peculiar property of those results is that they have been systematically developed in a single coherent framework originating from statistical mechanics. The results of Achkasov and Bakut and Ivanchuk have inspired many practical implementations.  Apparently these earlier results from the 1970s and 1980s in the Soviet literature were unknown outside the Soviet Union. The results were rediscovered in the West beginning in the late 1990s.  To this day, this earlier Soviet literature remains very poorly known and is almost never referenced in the modern literature.   

This project focusses on key results in the development of a theory of multi-object tracking and filtering with point processes from their roots in statistical mechanics through the Soviet literature. These form a coherent and sequential lineage in the chronology of methodological development of the theory of point processes, their representations, conditional estimation, and estimation of signals from a sequence of signals.  (In the modern Western literature, signals are called targets and their estimates and called tracks.)  

Daniel Clark, University of Southampton, UK

Alexey Narykov, University of Liverpool, UK

Roy Streit, Metron, USA

The project

 

The project is based on a series of journal articles originally published in Russian and (mostly) translated into English from 1963-1979.  These form a coherent development of the subject in the Soviet Union and cover essential topics in information and statistical estimation for multi-object systems.

Our view is that these works lay the foundation for modern multi-target tracking systems based on point processes. However, they are almost unknown in the tracking community. Hence our goal is to communicate these results to help advance the state of the art in multi-object filtering.

To ensure coherency of the project, the editors will write a foreword explaining the historical context of the works and curate a collection of commentaries written by different experts in the field of multi-object estimation.

In this site we present an overview of the selected articles in the development of the theory in chronological order and have categorised them into some topics that are familiar to researchers working in the field. These articles have been selected by tracing back the lineage from the citations in the later works through to the earlier foundation.  The project will demonstrates the development and context of the work, aiming to present a concise set of commentaries that are authentic to the original contributions.