Probabilistic reasoning in intelligent vehicle highway systems

This dissertation examines the task of situation assessment based on uncertain and incomplete sensor observations, and it argues that using Bayesian probability theory to reason about uncertainty yields methods that are not only sound in theory but also effective in practice. The main contributions of this dissertation include a theoretical framework for reasoning about object identification, new heuristic methods for handling intractable problems in uncertain reasoning, and several successfully implemented prototype systems that apply Bayesian models (such as dynamic probabilistic networks) to problems in the domain of intelligent highway vehicle systems.; Object identification is the task of deciding that two observed objects are in fact one and the same object. This dissertation describes the patterns of reasoning that allow object identity, which is essentially a first-order concept in predicate logic, to be grounded in raw sensory observations, which are essentially propositional. The theory is applied in a system that recognizes cars observed by cameras at widely separated sites in a freeway network. Using novel heuristics to handle the inevitable complexity of matching large numbers of objects, the system achieves high levels of performance.; Higher-level tasks in situation assessment require rich models for representation and reasoning. This dissertation explores how dynamic probabilistic networks provide a flexible, theoretically sound framework for encoding knowledge about specific domains, how they integrate low-level observations accumulated over time, and how they can provide high-level, symbolic assessments of situations. The theory is applied in two systems: one that performs symbolic traffic scene analysis on a fixed section of a freeway network, and another that monitors the environment of an autonomously controlled vehicle.