| This thesis addresses the estimation of highway vehicle travel times and origin-destination counts from data obtained by two or more roadside sensors. This research is motivated by a rich data set---as well as the promise of future data---in which detailed, multivariate observations are made on individual vehicles. With such detailed data, we attempt to probabilistically associate observations from different sensors as belonging to the same physical vehicle.;Chapter 1 describes the nature of our estimation problem, as well as some of the issues encountered in real applications. We also describe the motivating data.;In Chapter 2 we build a probability model that includes entrances and exits, sensor failures, and vehicles arriving at a sensor outside the monitoring period. A Monte Carlo EM algorithm is the primary tool for fitting the model and making inference on our quantities of interest. In this chapter we also discuss methods for selecting variables and for making the best use of high-dimensional sensor observations.;In Chapter 3 we apply our methods to simulated data and the motivating data of Chapter 1. Some heuristics for practical implementation are also discussed. |
