Sensor Data Fusion for Context-Aware Computing

Sensor Data Fusion for Context-Aware Computing Research

Summary of Contributions

Introducing Dempster-Shafer theory to context-aware computing

It is suitable to combine objective data with subjective judgments
Add uncertainty management -- able to handle ambiguity and ignorance
The reasoning process accounts for all evidence as an ensemble -- it can handle nested hypotheses that cannot be handled by the classical Bayesian methods
When hypotheses are mutually exclusive, the canonical Bayesian method emerges clearly as a subset of DS: bottom-line DS is as good as Bayesian-based methods

Extending Dempster-Shafer theory

Easily realize differential trust scheme on sensors, which cannot easily be handled in traditional sensor fusion methods, and it provides a easy way for human intervention
Mitigate conflicts that cause counter-intuitive (to some people) results using the classic Dempster-Shafer evidence combination rule
Incorporate sensors' behavior evolution (drift) information, thus outperform traditional methods that are static

System-building methodology and context-sensing architecture

Context consolidated, context-requiring applications are further separated from context-sensing implementation
Sensor observation joint distribution not required (DS-related)
Robust to change in sensor set and sensors' characteristics
The system is modularized to allow substitution of existing, or addition of new, sensor fusion or AI algorithms

Ph.D. Dissertation Abstract

Sensor Data Fusion for Context-aware Computing Using Dempster-Shafer Theory
(a copy of the dissertation in PDF format is avalaible here)

Towards having computers understand human users' "context" information, this dissertation proposes a systematic context-sensing implementation methodology that can easily combine sensor outputs with subjective judgments. The feasibility of this idea is demonstrated via a meeting-participant's focus-of-attention analysis case study with several simulated sensors using prerecorded experimental data and artificially generated sensor outputs distributed over a LAN network.

The methodology advocates a top-down approach: (1) For a given application, a context information structure is defined; all lower-level sensor fusion is done locally. (2) Using the context information architecture as a guide, a context sensing system with layered and modularized structure is developed using the Georgia Tech Context Toolkit system, enhanced with sensor fusion modules, as its building-blocks. (3) Higher-level context outputs are combined through "sensor fusion mediator" widgets, and the results populate the context database.

The key contribution of this thesis is introducing the Dempster-Shafer theory of evidence as a generalizable sensor fusion solution to overcome the typical context-sensing difficulties, wherein some of the available information items are subjective, sensor observations' probability (objective chance) distribution is not known accurately, and the sensor set is dynamic in content and configuration. In the sensor fusion implementation, this method is further extended in two directions: (1) weight factors are introduced to adjust each sensor's voting influence, thus providing an "objective" sensor performance justification; and (2) when the ground truth becomes available, it is used to dynamically adjust the sensors' voting weights. The effectiveness of the improved Dempster-Shafer method is demonstrated with both the prerecorded experimental data and the simulated data.

Presentation

My thesis research recieved funding award from Motorola University Partnership in Research (UPR) program. The following picture links to the presentation I gave to the Motorola Science Advisory Board Associates (SABA) annual meeting of 2002, held on October 19, in Dallas, Texas. (This is converted from Microsoft Powerpoint presentation file, you may need the newer version of Microsoft Internet Explorer to view it with best effects, you may need several clicks on the slides to view animated stuff.)