Citation

Formosa N, Quddus M, Ison S, Abdel-Aty M, Yuan J. Accid. Anal. Prev. 2020; 136: e105429.

Affiliation

Department of Civil, Environmental and Construction Engineering, University of Central Florida, Orlando, FL, 32816, USA. Electronic address: jinghuiyuan@knights.ucf.edu.

Copyright

(Copyright © 2020, Elsevier Publishing)

DOI

10.1016/j.aap.2019.105429

PMID

31931409

Abstract

Recently, technologies for predicting traffic conflicts in real-time have been gaining momentum due to their proactive nature of application and the growing implementation of ADAS technology in intelligent vehicles. In ADAS, machine learning classifiers are utilised to predict potential traffic conflicts by analysing data from in-vehicle sensors. In most cases, a condition is classified as a traffic conflict when a safety surrogate (e.g. time-to-collision, TTC) crosses a pre-defined threshold. This approach, however, largely ignores other factors that influence traffic conflicts such as speed variance, traffic density, speed and weather conditions. Considering all these factors in detecting traffic conflicts is rather complex as it requires an integration and mining of heterodox data, the unavailability of traffic conflicts and conflict prediction models capable of extracting meaningful and accurate information in a timely manner. In addition, the model has to effectively handle large imbalanced data. To overcome these limitations, this paper presents a centralised digital architecture and employs a Deep Learning methodology to predict traffic conflicts. Highly disaggregated traffic data and in-vehicle sensors data from an instrumented vehicle are collected from a section of the UK M1 motorway to build the model. Traffic conflicts are identified by a Regional-Convolution Neural Network (R-CNN) model which detects lane markings and tracks vehicles from images captured by a single front-facing camera. This data is then integrated with traffic variables and calculated safety surrogate measures (SSMs) via a centralised digital architecture to develop a series of Deep Neural Network (DNN) models to predict these traffic conflicts. The results indicate that TTC, as expected, varies by speed, weather and traffic density and the best DNN model provides an accuracy of 94% making it reliable to employ in ADAS technology as proactive safety management strategies. Furthermore, by exchanging this traffic conflict awareness data, connected vehicles (CVs) can mitigate the risk of traffic collisions.

Copyright © 2020 Elsevier Ltd. All rights reserved.

Language: en

Keywords

Deep Neural Network (DNN); Regional–Convolution Neural Network (R-CNN); Safety Surrogate Measures; data integration architecture; traffic conflicts