Developing public transport information systems with use of smartphone-based human probe data

Abstract

This thesis contributes to the development of public transport information systems with use of human probe data. New methods for estimating three key performance indicators (KPIs) of bus transit systems are proposed in this study; namely, average bus passenger waiting times, real-time bus arrival times, and bus crowding levels. Without the data availability from in-vehicle sensing devices, smartphone-based human probe data are considered as the potential alternative data sources for estimating the three KPIs of bus transit system. This thesis firstly focuses on the non-participatory sensing approach, passive Wi-Fi data. Through a detailed investigation, this thesis aims to understand the capabilities and limitations of using passive Wi-Fi data for deriving the bus transit information. The thesis proposes a new method for estimating the average bus passenger waiting times at a single bus stop. The proposed method is designed to handle massive noise and the temporal uncertainties of these data. In order to achieve this, generalized classification features are introduced for describing the attributes of individual Wi-Fi devices at a bus stop. The features can then be used for identifying waiting passengers' devices and estimating the average bus passenger waiting times at bus stop. The thesis then proposes a novel framework for developing a real-time bus arrival time information system using participatory-based bus data contributed by bus passengers. This real-time information can be provided without the need for bus tracking devices and data provision from the bus operators. The proposed framework is developed to cope with the particular characteristics of participatory-based bus data such as inconsistencies in the bus data due to the participation of multiple passengers on the same bus, and the availability of bus data in the spatial and temporal dimensions. Finally, the participatory-based bus data are used for developing a bus crowding prediction system. This system aims to predict the bus crowding levels of individual buses when they arrive at each bus stop. The crowding levels can be provided together with the bus arrival times so that passengers can make their boarding decisions with more relevant information when planning their journeys. Data mining techniques are adopted for the prediction of bus crowding levels based on several relevant factors, i.e. bus dwell times and bus headways at each bus stop. In the long run, the system developed in this thesis will be able to perform bus crowding prediction without the availability of passenger boarding and alighting information at bus stops.