School of Civil Engineering

Leveraging Urban Trajectory Data to Enhance Network-Wide Real-Time Traffic Flow Prediction

Recent advances in tracking technologies and the ubiquitous presence of location-aware sensors in urban spaces (e.g., GPS, WiFi, Bluetooth, and RFID devices) have produced large amounts of urban trajectory data, capturing the detailed spatiotemporal footprint of the movement of individual people and vehicles. The increasing availability of trajectory data is providing new opportunities for transportation researchers and practitioners to develop data-driven models that can leverage insights into detailed movements of vehicles and people. Despite the potential benefits of urban trajectory data as new sources of insights and additional information to enhance transport modeling, the current network modeling and management practices still rely heavily on conventional fixed-point traffic data (i.e., loop detector data) and do not exploit the full potential of these emerging trajectory datasets.

This project aims to develop an analytical framework for leveraging urban trajectory data to enhance the prediction of traffic flows for network-wide real-time traffic management. More specifically, this project aims to address the following research gaps in the existing literature: i) the existing studies in real-time traffic flow prediction focus on predicting link-level traffic condition, such as link flow and link speed, and do not fully capture network-wide traffic flow dynamics (the number of vehicles that travels from one region to another); ii) the existing studies mainly use fixed-point traffic data collected from loop detectors and how the incorporation of new trajectory data can improve the prediction performance is largely unstudied; and iii) the existing studies that attempted to use trajectory data for traffic prediction problems do not fully address the challenges that trajectory datasets do not represent the whole vehicle population, but only cover a sample of vehicles.

To fill the above-mentioned research gaps, this project proposes specific tasks that involve: i) the development of effective data-driven models that can learn network-wide mobility patterns embedded in large-scale trajectory data and predict the flow between regions in real-time; ii) in-depth discussion for comparing urban trajectory data with fixed-point traffic data to mathematically and emprically show the usefulness of trajectory data in traffic prediction applications; and iii) the investigation of the effect of trajectory data coverage on prediction performance and methods to infer the population movement patterns from a sample of trajectories.

1:00pm - 2:00pm

Thursday, 20 December 2018


AEB Boardroom 601
Advanced Engineering Building 49