Sensor fusion: a rough granular approach
James Francis Peters, Sheela Ramanna, Andrzej Skowron, Jarosław Stepaniuk, Zbigniew Suraj · 2002
The paper introduces an application of a particular form of rough granular computing in fusing (combining) sensor readings. The intent of the paper is to describe a system that engages in a form of knowledge discovery based on sensor fusion. Such a system responds to sensor outputs in a manner that is selective, determines the relevance of each sensor in a classification effort, and constructs information granules computationally useful in arriving at a decision (proposed solution to a problem) in a problem-solving system. A sensor is a device that responds to each stimulus by converting its measured input to some form of usable output. Relevance of a sensor is computed with a rough integral that computes a form of ordered weighted average of sensor values. The construction of an information granule depends on the selection of a threshold for sensor values. Only those sensors with rough integral values approaching a selected threshold are fused (i.e., used to construct a granule). The contribution of the paper is the introduction of a sensor fusion method based on rough integration. By way of practical application, an approach to fusion of homogeneous sensors deemed relevant in a classification effort is considered.