Real-time crash risk estimation using machine learning

The aim of this PhD project is to develop next-generation models for crash risk prediction in real-time. Leveraging the combined benefits of Intelligent Transport Systems and computer-vision techniques, traffic conflicts – which represent temporal/spatial proximity of various road users including vehicles, pedestrians and bicyclists – will be obtained from video recordings at signalised intersections in Queensland.

By combining the state-of-the-art machine learning algorithms, this study introduces a novel modelling approach that can predict traffic crashes in real-time from traffic conflicts with a high degree of accuracy. The use of traffic conflict data for crash risk estimation is itself a new paradigm.

The proposed approach will be rigorously tested and compared with existing state-of-the-art methods like Extreme Value Theory.

Finally, the proposed approach will be employed to estimate crash risk in real-time at a signal cycle level and identify crash-prone conditions that contribute to crash occurrence. Moreover, crash mitigation strategies shall also be developed.

This project will contribute to both methodological and theoretical advancement in crash risk modelling. The findings of this project will be applied to develop an Artificial intelligence (AI)-based analytics tool that will help Transport Management Centres and state road authorities to make traffic management decisions in real-time.

Participants

• Queensland University of Technology

Project background

Existing traffic safety assessment follows a reactive approach that is predominantly based on police-reported crash data, which suffer from several issues such as under-reporting, low sample size, etc. Further, the current reactive road safety assessment is not capable of assessing real-time risks on road users, and as such, the real-time risk mitigation strategies could not be developed.

Motivated by this research need, in this PhD project, next-generation models for real-time crash risk prediction shall be developed utilising Artificial intelligence techniques.

The proposed models shall be rigorously tested and compared with existing state-of-the-art methods like Extreme Value Theory.

Project objectives

The objective of this project is to develop a real-time crash prediction model using Artificial Intelligence and Computer Vision techniques.

It is intended that the model developed will enable crash mitigation strategies to be developed which are intended to aid Transport Management Centres and state road authorities to make traffic management decisions in real-time.

Please note …

This page will be a living record of this project. As it matures, hits milestones, etc., we’ll continue to add information, links, images, interviews and more. Watch this space!