Traffic state estimation in the absence of dedicated infrastructure using crowd-sourced trajectory data

An essential part of regulating and optimising traffic flow in urban and highway networks is Traffic state estimation (TSE). It provides information on road-traffic conditions that may be utilised to improve traffic flow, optimize the performance of the transportation system, support the development of Intelligent Transportation Systems (ITS), revolutionise the transport industry and raise the quality of life for those living in cities and suburbs. TSE needs dedicated expensive infrastructure for transportation data collection, depending on a city’s transport infrastructure and the level of economic development. Many Low- and Middle-Income Cities (LMIC) do not have well-established transportation infrastructure. As a result, there is limited availability of expensive sensor data for estimating the traffic state. Therefore, there is a need to develop more cost-effective and efficient methods for estimating traffic states in the absence of dedicated sensor-based data.
Consequently, recent studies have shown that trajectory data collected from vehicles equipped with Global Navigation Satellite Systems (GNSS) units can be used for TSE reducing the dependency on dedicated sensor data. Trajectory data are also combined with static sensor data to estimate relevant parameters of traffic states, such as traffic flow, travel times, queue lengths, and shockwave boundaries. Commercial trajectory data are highly privacy-sensitive and not always accessible in a location of interest. So anonymized crowd-sourcing has been used in some studies to collect trajectory data without hindering privacy issues.
Crowd-sourced trajectory data at least contain information on location, time, and unique anonymous ID in discrete time steps. However, crowd-sourced trajectory data collection methods bring several challenges due to various systematic and device errors, and a limited number of trajectories collected with an unknown fraction of the population traffic. Also, there are inherent difficulties with the existing trajectory-based research to be applied for TSE. Crowd-sourced trajectory data points do not contain essential traffic information, such as what ground truth route was taken on a road network, whether the data was collected from a moving vehicle on road traffic, and how it interacted with traffic conditions. These challenges with crowd-sourced trajectory data emphasise the necessity for novel methods to get around the current limitations of estimating traffic state parameters.
This thesis proposes new techniques to utilize crowd-sourced trajectory data in order to estimate traffic flow, density, velocity, and travel time as traffic state parameters of a road network. An overall hypothesis of the research problem, a set of research questions and their corresponding research goals are developed by analyzing research gaps in crowd-sourced trajectory data-based TSE. The challenges with data collection methods and limitations on the applicability of the existing methods are established in the individual research questions. The first two research goals estimate ground truth routes of vehicles on the road by an improved map-matching and travel mode detection of the multi-modal raw trajectory data. The remaining research goals involve developing methodologies for estimating parking information, traffic counts, and vehicle Origin-Destination (OD) flows from vehicle-bound trajectories. Finally, the estimated traffic counts, and OD flow, together infer travel time, traffic flow, density, and velocity at the road network of interest.
Thus, this thesis contributes to TSE only using crowd-sourced trajectory data at the road network of interest where dedicated data collection infrastructure does not exist or is not operational.
The thesis draws upon empirical data and real-world case studies to validate and evaluate the proposed methods and models. It also discusses the limitations and potential applications of the findings, highlighting the implications for traffic management, urban planning, and transportation systems. All the proposed methods within the research goals demonstrated usefulness including the future applicability of crowd-sourced trajectory data as a low-cost solution for TSE in LMIC. This thesis serves as a valuable contribution to the field of traffic management systems by demonstrating the feasibility and benefits of using crowd-sourced trajectory data in traffic estimation. It contributes to a better understanding of urban traffic conditions in LMIC, where road infrastructure is inadequate and resources are limited. The proposed techniques can help in optimizing traffic operations and planning, which can result in economic and ecological loss control. Ove all, this research has the potential to make significant contributions to the field of transportation engineering and improve the quality of life in LMIC.