Efficient tracking of public transit system in urban environment

Public transport is an essential component of any economy. It reduces traffic on the road and permits free flow of humans and goods. Advanced Traveler Information Systems (ATIS) are used in Intelligent Transportation Systems to provide real-time transit information concerning the location of vehicles and their predicted arrival times. Such information is computed at base stations using data transmitted from on-board sensors. The objective of ATIS is to improve the quality of service for the commuters and make public transit as their primary mode of transportation.;In this dissertation, we develop algorithms to improve the tracking accuracy of ATIS in a communication constrained environment. First, we develop novel stochastic models for predicting the motion of vehicles on interconnected road segments, each of which has different traffic characteristics. The resulting model is a hybrid system, with state dependent transitions as vehicles traverse from segment to segment. These models incorporate side information such as average speed on a segment of the road, quality of GPS estimation, desired path and road interconnectivity. We develop new estimation algorithms that incorporate the state-dependent switching among models by using adaptive predictors to estimate switching times. We also develop robust estimation algorithms to reduce the effect of outlier measurements, created by potential GPS anomalies. In particular, we present an outlier rejection procedure for particle filters that uses theory from kernel density estimation and probability level sets. The simulation examples show that an improvement of more than 25% can be achieved with the proposed hybrid models compared to the homogeneous model of representing each road with the same model. Further, the outlier rejection algorithm for particle filter can improve the tracking accuracy by 19% compared to the filters that do not employ the rejection algorithm.;In the last part we propose processing and communication architectures for ATIS systems operating in a communication-constrained environment. We consider architectures where the choice of communication times is made adaptively at the vehicles. For each of these architectures, we develop adaptive algorithms to determine when vehicles should communicate their position to the base station, or when the base station should request a position update from the vehicle. Depending on the traffic conditions and quality of the predictor, simulation examples show that communication costs can be reduced by almost 75% with hardly any loss in accuracy.;The research in this dissertation establishes that ATIS can be implemented in communication constrained environments and improved performance can be achieved through the use of hybrid system motion models and robust estimation techniques based on these models. This work also lays foundation for several new research directions including ATIS in a probabilistic environment and sophisticated hybrid models.