Combinatorial structures and filter design in information spaces

Yu, Jingjin

Permalink

https://hdl.handle.net/2142/44500 Copy

Description

Title

Combinatorial structures and filter design in information spaces

Author(s)

Yu, Jingjin

Issue Date

2013-05-24T22:18:24Z

Director of Research (if dissertation) or Advisor (if thesis)

LaValle, Steven M.

Doctoral Committee Chair(s)

LaValle, Steven M.

Committee Member(s)

• Hutchinson, Seth A.
• Liberzon, Daniel M.
• Mitra, Sayan

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Combinatorial Filter
• Shadow Information Spaces
• Sensor Fusion
• Situation Awareness

Abstract

In this thesis, we develop a filtering process called combinatorial filters for handling combinatorial processes that evolve over time and study two practical problems using this method. The first problem is a generalization of the sensing aspect of visibility-based pursuit evasion games, in which the task is to maintain the distribution of hidden targets that move outside the field of view while a sensor sweep is being performed. For this problem, we apply information space concepts to significantly reduce the general complexity so that information is processed only when the shadow region (all points invisible to the sensors) changes combinatorially or targets pass in and out of the field of view. The cases of distinguishable, partially distinguishable, and completely indistinguishable targets are handled. Depending on whether the targets move nondeterministically or probabilistically, more specific classes of problems are formulated. For each case, efficient filtering algorithms are introduced, implemented, and demonstrated that provide critical information for tasks such as counting, herding, pursuit-evasion, and situational awareness. Next, we study the problem of using sparse, heterogeneous sensor data to verify the stories (i.e., path samples) of agents. Since there are two sets of data, the combinatorial filter for this problem can be built in two ways: Using a filter (an automaton) built from sensor data to process the story or using a filter built from the story to process the sensor data. Both approaches lead to dynamic programming based efficient algorithms for extracting a compatible path if one exists. In addition to exact path inference, our method also applies to approximate path inference that allows errors in data. Besides immediate applicability toward security and forensics problems, the idea of behavior validation using external sensors also appears promising in complementing design time model verification.

Graduation Semester

2013-05

Permalink

http://hdl.handle.net/2142/44500

Copyright and License Information

Copyright 2013 Jingjin Yu

Owning Collections