Mahler R.P.S. Advances in Statistical Multisource-Multitarget Information Fusion

Boston; London: Artech House, 2014. — 1167 p. — ISBN13: 9781608077984, 1608077985.This book is a sequel to 2007 book, Statistical Multisource-Multitarget Information Fusion [179]. That earlier book was a textbook-style introduction to finite set statistics (also known as random set information fusion), a fundamentally new, seamlessly unified, and fully probabilistic approach to multisource-multitarget detection, tracking, classification, and information fusion.
This sequel provides a comprehensive description of the state of the art in random set information fusion since 2007—a description not otherwise available.
Its intended audience is signal processing graduate students, researchers, and engineers, as well as mathematicians and statisticians interested in tracking, information fusion, robotics, and related subjects.
Finite-set statistics has five major elements:
A general theory of measurements, based on a stochastic-geometry formulation of random set theory.
A general theory of stochastic multiobject systems, based on a stochasticgeometry formulation of point process theory or, equivalently, random finite set theory.
A general approach to multisource-multitarget modeling based on multiobject integro-differential calculus.
A general optimal approach to multisource-multitarget processing based on these models and Bayesian filter theory.
A general approach to multisource-multitarget algorithmic approximation, also based on multiobject integro-differential calculus.
Since 2007, the approach has inspired a considerable amount of research, conducted by many dozens of researchers in at least a dozen nations, reported in many hundreds of research publications. As a result, progress in random set
information fusion has been rapid and has proceeded in diverse and sometimes unexpected directions, propelled by many clever new ideas. Indeed, the rapidity and extent of progress has itself been somewhat unexpected, especially given cautious disclaimer in Statistical Multisource-Multitarget Information Fusion (/file/949669/, p. 566):
“...preliminary research has suggested that the PHD and CPHD filters may be more effective than [multihypothesis correlator-type] filters in some conventional multitarget detection and tracking problems. Whether such claims hold up is for future research to determine. Here we emphasize that the PHD and CPHD approaches were originally devised to address non-traditional tracking problems such as those just described [that is, tracking of target clusters].”
A summary of advances in the field will be given in Section 1.2. In brief, the progression of research emphasis has been roughly as follows: from PHD filter to CPHD filter; from CPHD filter to multi-Bernoulli filters and, in particular, to the CBMeMBer filter; and most recently, to “background-agnosic” CPHD and CBMeMBer filters and the Vo-Vo exact closed-form multitarget detection and tracking filter. Ancillary advances have occurred in regard to joint tracking and sensor registration; superpositional sensors and track-before-detect (TBD); distributed fusion; sensor management; and robotics.
As time progressed, it became increasingly clear that the most intriguing aspects of the new research should be aggregated and systematized, in a single place, into a coherent and integrated picture. That is the purpose of this book. Thus one of primary goals is to provide a deep-dive overview of the state of the art in the field.