Contribution to multisensor fusion formalization

Abstract Using data fusion for scene analysis to take advantage of the capabilities of each sensor, and palliate their limitations, is very attractive. However, data fusion involves new problems such as control of increasing data flow, knowledge modeling, or strategy and reasoning determination. Controlling these problems is the only way to obtain all the potential advantages of data fusion. Thus, in Part 2, we focus on sensor relationships. We propose a quantitative measure or redundancy and complementarity between various types of sensors. For this, we use information theory, introduced by Shannon, that gives a very interesting mathematical framework. In Part 3, we focus on data fusion mechanisms. We give a brief review of multisensor fusion methods that can be found in the literature, and we indicate the few attempts at classification we found. These classifications usually describe only some characteristics of a multisensor fusion process. In order to better formalize these characteristics, we define a generic fusion cell that can describe any fusion process. Integration is viewed as an association of different fusion cells. This formalization gives us a more generic representation of the integration problems, while splitting off conceptually the fusion problems from the integration problems.