Modeling and fusing uncertain multi‐sensory data
O. A. Basir, H. C. Shen · Journal of Robotic Systems · 1996
To develop a Multi-Sensor System (MSS) one has to consider the following three important issues: (1) modeling the uncertainty that exists in the sensory measurements, (2) modeling the cooperation behavior among the sensors, and, finally, (3) developing fusion strategies that recognize both the uncertainty model and the cooperation behavior. In this article we propose a probabilistic approach for modeling the uncertainty and cooperation in sensory teams. We show how the Information Variation measure can be used to capture both the quality of sensory data and the interdependence relationships that might exist between the different sensors. This allows the sensor fusion procedures to avoid the assumption that the observations made by the different sensors are totally independent, an assumption that lessens the applicability of such procedures in many practical situations. We also show how DeGroot's Consensus model can be combined with the Information Variation model to fuse the uncertain sensory data. The proposed approach develops to an approximation of the Bayesian paradigm when the team constitutes more than two sensors. It is the computational burden as well as the difficulty associated with the construction of the exact Bayesian paradigm that motivated the development of this approach. © 1996 John Wiley & Sons, Inc.