Real-time detection of crash-prone conditions at freeway high-crash locations

John N. Hourdos, Vishnu Garg, Panos G. Michalopoulos, Gary A. Davis

Research output: Chapter in Book/Report/Conference proceedingConference contribution

60 Scopus citations


Because of growing concern over traffic safety and rising congestion costs, recent research efforts have been redirected from the traditional reactive traffic management (crash detection and clearance) toward on-line proactive solutions for crash prevention. Such a solution for high-crash areas is explored by the identification of the most relevant real-time traffic metrics and the incorporation of them in a model to estimate crash likelihood. Unlike earlier attempts, this model is based on a unique detection and surveillance infrastructure deployed on the freeway section that has the highest crash rate in Minnesota. State-of-the-art infrastructure allowed the video capture of 110 live crashes, crash-related traffic events, and contributing factors while measuring traffic variables (e.g., individual vehicle speeds and headways) over each lane in several places in the study area. This crash-rich database was combined with visual observations and analyzed extensively to identify the most relevant real-time traffic measurements for detecting and developing an on-line model of crash-prone conditions. This model successfully establishes a relationship between quickly evolving real-time traffic conditions and crash likelihood. Testing was performed in real time during 10 days not previously used in model development, under varied weather and traffic conditions. The crash likelihood model - and, in turn, the detection algorithm - succeeded in detecting 58% of the crashes, with a 6.8 false decision rate.

Original languageEnglish (US)
Title of host publicationArtificial Intelligence and Advanced Computing Applications
PublisherNational Research Council
Number of pages9
ISBN (Print)0309099773, 9780309099776
StatePublished - 2006


Dive into the research topics of 'Real-time detection of crash-prone conditions at freeway high-crash locations'. Together they form a unique fingerprint.

Cite this