Multi-Sensor Data Fusion in Presence of Uncertainty and Inconsistency in Data
Author: Manish Kumar
Publisher:
Published: 2009
Total Pages:
ISBN-13: 9783902613523
DOWNLOAD EBOOKSensors measurements are inherently uncertain and often inconsistent. Appropriate consideration of uncertainty and identification/elimination of inconsistent measurements are essential for carrying out accurate estimation. The research reported in this chapter proposes a unified and formalized approach to fuse data from multiple sources which can take uncertainty of sensor data into account and automatically identify inconsistency in sensor data. Appropriate modeling of uncertainties in sensor measurement is necessary. This chapter presents an innovative neural network based method to model sensor's uncertainties. Further, the chapter presents a strategy that adds a term to the popular Bayesian approach corresponding to a belief that the sensor data is not spurious conditioned upon the data and true state. An information theoretic measure is utilized to observe the information content of the posterior distribution to identify and eliminate inconsistent data. An extensive simulation study was performed where data from three sensors was fused. It was observed that the presented method was very effective in identifying spurious data, and, elimination of spurious data ensured more accurate results. Finally, the effectiveness of the proposed technique to identify and eliminate inconsistent sensor data in sequential Bayesian fusion was demonstrated with the help of an experiment performed in a robotic workcell where measurements from stereo vision, infra-red proximity, and laser proximity sensor were fused to obtain three-dimensional occupancy profile of robotic workspace.
