Growing concern over traffic safety as well as rising congestion costs have been recently redirecting research effort from the traditional crash detection and clearance reactive traffic management towards online, proactive crash prevention solutions. In this project such a solution, specifically for high crash areas, is explored by identifying the most relevant real time traffic metrics and incorporating them in a crash likelihood estimation model. Unlike earlier attempts, this one is based on a unique detection and surveillance infrastructure deployed on the freeway section experiencing the highest crash rate in the state of Minnesota. This state-of-the-art infrastructure allowed video recording of 110 live crashes, crash related traffic events, as well as contributing factors while simultaneously measuring traffic variables such as individual vehicle speeds and headways over each lane in several places inside 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 crash prone conditions and develop an online crash prone conditions model. This model successfully established a relationship between fast evolving real time traffic conditions and the likelihood of a crash. Testing was performed in real time during 10 days not previously used in the model development, under varying weather and traffic conditions. (Author/publisher)
Abstract