Tracking and data fusion: Handbook of Algorithms (Bar-Shalom, Y., et al; 2011) [Book Review

Abstract

Three professors from the University of Connecticut have given us a new book that is both broad and deep, both practical and theoretical, and both serious and amusing, laden with algorithms, plots, equations, and wisdom. This book will help hard-boiled engineers looking for fast practical algorithms that work robustly in the real world. It will also enlighten academics about important and interesting topics yearning for even more theoretical innovations. The only problem is actually reading this everlasting gobstopper of a book (with 1235 pages, 2900 equations, 470 figures, 130 tables, hundreds of references, too many footnotes to count, and more). My own approach to solve this problem is to start at the back of the book to see how it ends. There we find a very long list of references, but most important we want to see how many of our papers the authors have cited. This is followed by a careful study of the footnotes, which are a cornucopia of jokes (as well as many serious technical comments). At last, we dive into hundreds of pages of detailed algorithms with numerical results, garnished with a stiff dose of stochastic matrix calculus. Neophytes are often surprised that 1235 pages are required to cover the subject of tracking and multisensor data fusion, considering that there are only 19 crucial issues to be considered, including limited resolution of the sensors, data association errors, measurement noise errors of the sensors, residual sensor bias and drift errors, measurement errors due to the propagation through the physical environment (e.g., the troposphere and ionosphere for radar and underwater for sonar), clutter, jamming, multipath, nonzero probability of false alarms, nonunity probability of detection of the targets, ill-conditioning of the error covariance matrix of the extended Kalman filter (or other algorithm), nonlinearities in the sensors measurements and dynamics of the targets, unmodeled dynamics of the target motion, real-time computational complexity of the algorithms, real-time joint multiple sensor resource management, limited resolution of the sensors, an ever-expanding plethora of algorithms invented by engineers and academics, non-Gaussian multimodal probability densities of the estimate of the state vector and the sensor measurements, statistical inconsistency of covariance matrices of the sensor measurements and the state vector error covariance matrices from various sensors, random sets of hypotheses, robustness to uncertainty in the models of sensors and target dynamics and environment, quantification of system performance as a function of track rate, multiple sensor geometry, unmodeled target acceleration, signal-to-noise ratio, and spatial density of the targets, as well as the most important issue, limited resolution of the sensors.