[ main page ] [ back ]

2006 : Extended Confidence-Weighted Averaging in Sensor Fusion

Author(s)
Angela Sch�rgendorfer
Abstract
Sensor fusion refers to the practice of combining information gathered by multiple sensors to receive a more accurate, dependable and comprehensive image of a system’s environment. The need for such strategies arises from the fact that single sensors have limited dependability since they are subject to environmental interference or hardware noise, as well as possible sensor failure. A system depending on sensor information should therefore not rely on the information from a single sensor.
This thesis discusses the currently established techniques for sensor fusion, focussing on methods for merging raw sensor data. Analyzing the various shortcomings of established methods, a new method for stateless fusion of raw sensor data is presented—the technique of extended confidence-weighted averaging. It is based on the previously suggested confidence-weighted averaging method, but extends it by including known correlations between the error distributions of the sensors. The benefits expected from this new method are an increase in the accuracy of the fusion result as well as a more dependable estimate of the uncertainty associated to the fused result.
An evaluation on real sensor data collected using a mobile robot analyzes the improvements achieved by the newly presented method.
Bibtex
@mastersthesis{ schörgendorfer:2006,
  author =      "Angela Schörgendorfer",
  title =       "Extended Confidence-Weighted Averaging in Sensor Fusion",
  address =     "Treitlstr. 3/3/182-1, 1040 Vienna, Austria",
  school =      "Technische Universit{\"a}t Wien, Institut f{\"u}r Technische Informatik",
  year =        "2006"
}
Download


[ main page ] [ back ]