A Sensor Fusion System Using Physical Network As Internal Models

Publisher Website
Google Scholar
Abstract
Sensor fusion is basic structure of systems with various sensors and actuators to implement functions which can not be performed with a few sensors. In this paper, a general definition of sensor fusion, which can clarify problems of sensor fusion, is proposed. The definition shows that the sensor fusion is conversion into a stable state with consistency. In order to implement the sensor fusion, it is the most significant problem to harmonize requirements of high speed and real time processing and abstractive knowledge processing.From the viewpoint of this definition, a paradigm of internal model is also proposed. The internal model is an objective model constructed in processing units. It can naturally describe a large quantity of polymodal sensors and actuators so that it can implement flexibility and robustness into the fusion system. In addition, a system, in which realtime signal processing and complicated symbol processing are separated utilizing the paradigm, is also proposed. The signal processing subsystem of the system is constructed with Transputers, which are processors to realize MIMD type parallel processing.The configuration of the system is described and sensor fusion between vision and pseudo tactile information is demonstrated as a practical sample of the system. Experimental results of sensor data fusion and applicaion to error detection and correction are shown.

This publication has 0 references indexed in Scilit: