Uncertainty management in multisensor robotic tracking systems
Abstract
As the application areas for robotic systems become more diverse and more complex, the problem of identifying, propagating, and reducing the sources of uncertainties within robotic and other automation systems is becoming a critical design and implementation problem. A new paradigm in uncertainty management is proposed for multisensor robotic or other automation systems. The new paradigm is based on the Uncertainty Management Network (UMN) which decomposes a system model into a hierarchical tree-like network structure of basic processing nodes interconnected using information propagation channels. The inputs to a system are first modelled using Maximum-Uncertainty Possibility Distribution type fuzzy numbers which are propagated through the system model using fuzzy set operations and arithmetic to produce a final fuzzy number output for the system. Uncertainty calculus is performed at each internal processing node in order to provide uncertainty measures for all internal stages of the system which can be used for decision-making, for performance analysis and improvement, and for multisensor fusion, which is used for uncertainty reduction. The Uncertainty-Reductive Fusion Technique is proposed to fuse together competitive information provided by multiple sensors and/or redundant processing in order to produce a representative consensus result which is more confident than the individual unfused information. One possible application area of the UMN-based paradigm in uncertainty management is the problem of sensor-based target tracking, and a real-time, dual-camera tracking system is used as an experimental testbed for the methods outlined in this research. The bottleneck to any visual tracking system is the large amount of data which needs to be processed in order to determine the 3-dimensional pose (position and orientation) information about the target, and so a visual tracking algorithm based on scanlines is proposed in order to implement a very fast tracking system which is capable of tracking objects moving at high speeds (abrupt or shaking motion), with partial object occlusion, and with minimal data processing.
Degree
Ph.D.
Advisors
Lee, Purdue University.
Subject Area
Electrical engineering|Industrial engineering|Mechanical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.