A Critical Examination of Sensor Fusion

James L. Farrell

Abstract:	Although the concept of sensor integration is by no means new. much effort in this area to date has concentrated on mixing outputs after considerable processing has been performed (e.g.. "blending" of track files computed separately from data obtained via Kalman estimators fed by isolated sensor information). Often the prevailing wisdom has been that the information "can't" be integrated until these separate track files have been initiated. While the other extreme (i.e.. bringing raw data together while still in rf analog form. at different parts of the spectrum) is not advocated here. the pace of advances in signal processing technology warrants repeated assessment of true fusion potential. Specifically a worthwhile objective is to define. for new generations of signal processors. at what point in the chain of cascaded operations should blending of information be initiated. The present answer to that question. it is contended here. is not at output track files fed by separate sensor data. A central problem of separate processing is misfiling of reports: observed initial position of one object is compared with subsequent position of another. leading to a velocity estimate that belongs to neither. That problem grows. almost beyond reason. with high target density. This is particularly true when sensor responses to valid targets are accompanied by extraneous responses. while the sensors receive information that is incomplete and/or ambiguous as to source. Consider. for example. a FUR providing only line-of-sight (LOS) directional data from multiple "hot targets" (including cows). while radar reports include echoes from other extraneous objects (e.g.. rocks) . A self-evident prescription: bring data together ahead of the detection-validity decision logic. In the example just cited. only those objects responding to both FUR and radar would then be allowed entry into an estimation or tracking algorithm. One manuscript is not going to solve the vast scope of data fusion problems (especially as approached here). but an assessment of relevant functions and their implications is clearly warranted. To that end the operation of a generic dual mode subsystem (FUR and radar) is addressed. under conditions of measurement availability that facilitate some of the requisite mapping operations. Reference 1 provides a convenient conceptual subdivision into "macro-registration" and .. micro-registration" which will be adopted here: the relation of that work to the present scope is also discussed in the sections that follow.
Published in:	Proceedings of the 44th Annual Meeting of The Institute of Navigation (1988) June 21 - 23, 1988 U.S. Naval Academy Annapolis, Maryland
Pages:	38 - 45
Cite this article:	Farrell, James L., "A Critical Examination of Sensor Fusion," Proceedings of the 44th Annual Meeting of The Institute of Navigation (1988), Annapolis, Maryland, June 1988, pp. 38-45.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In