Multisensor Fusion
A Minimal Representation Framework

This work develops a general theory of minimal representation multisensor fusion and applies it in a series of experimental studies of sensor-based robot manipulation. A novel application of differential evolutionary computation is introduced to achieve practical and effective solutions.

The fusion of information from sensors with different physical characteristics, such as sights, touch and sound, enhances the understanding of our surroundings and provides the basis for planning, decision-making, and control of autonomous and intelligent machines. The minimal representation approach to multisensor fusion is based on the use of an information measure as a universal yardstick for fusion. Using models of sensor uncertainty, the representation size guides the integration of widely varying types of data and maximizes the information contributed to a consistent interpretation. In this work, a general theory of minimal representation multisensor fusion is developed and applied in a series of experimental studies of sensor-based robot manipulation. A novel application of differential evolutionary computation is introduced to achieve practical and effective solutions to this difficult computational problem.

Introduction to multisensor integration
multisensor data fusion
multisensor fusion in object recognition
minimal representation
environment and sensor models
minimal representation multisensor fusion and model selection
multisensor fusion search algorithms
applying the abstract framework to concrete problems
multisensor object recognition in two dimensions
multisensor object recognition in three dimensions
laboratory experiments
discussion of experimental results.