Principles of Modern Radar - Volume 2 1891121537

15.3 Multisensor Tracking 69115.3 MULTISENSOR TRACKINGSection 15.2 focuses on the challenges in measurement-to-track data association andtrack filtering that arise in multitarget scenarios, as well as techniques for mitigating thesechallenges. This section shifts the focus to the challenges (and mitigating techniques)associated with multisensor tracking, also often termed ‘sensor fusion’.Sensor fusion is fundamental to numerous radar applications for several reasons,including necessity, observability, capacity, and robustness. This section discusses therationale for fusion, the two main architectures, and challenges associated with both.Necessity is the most obvious reason for sensor fusion. Simply put, the field of regardof a radar may be smaller than the area of interest for a given application. When thisoccurs, multiple sensors are necessary to obtain required coverage. Hence, sensor fusionis often necessary to allow adequate coverage of a particular region.Even if the radar’s field of regard is sufficiently large, the radar may still have a limitedability to observe features of interest for objects within this region. Suppose, for example,that a radar must detect a small target maneuver. If the maneuver occurs orthogonally tothe radar’s location, then the radar’s angle measurements may not be sufficiently accurateto detect it, rendering the cross-range maneuver unobservable. However, that same smallmaneuver may be observable to another radar at a different location. If spatially-distributedradars are networked, they provide spatial diversity and improved observability to a varietyof object features.The potential for large numbers of targets is inherent in some radar applications. Infact, the number of targets may exceed the (hardware or software) capacity of a singleradar. If a single radar cannot adequately service all required functions (e.g., acquisition,tracking, discrimination) for all potential targets, then the obvious solution is to deploy anetworked suite of radars. If data are properly associated and fused, the networked systemcan address a larger capacity of threats than would be possible with any single contributingradar.Finally, networking and fusion produces a more robust system. A single radar maymalfunction or be damaged by an adversary. If radars are intelligently networked, thefailure of a single one does not lead to utter failure; the remaining radars can still performthe mission, albeit at a somewhat degraded level (since fewer total observations areavailable).15.3.1 Sensor Fusion ArchitecturesSensor fusion architectures are fundamentally defined by the type of data (i.e., measurementsversus tracks) being fused. Variations are possible (and five options are discussedin [7]), but the two most common fusion alternatives are measurement-level fusion andtrack-level fusion.Measurement-level fusion (also sometimes referred to as centralized fusion in academicliterature [7]), in which all measurements from all sensors are transmitted andassociated to create system tracks, is theoretically the best choice, since it makes use ofall available data. However, constraints imposed by communications systems often renderthis impractical. Consider, for example, a notional set of radars that collect measurementson thousands of objects several times per second. Transmitting all of these measurementsfrom each radar to the rest of the network would require far more bandwidth than istypically available.