University of Minnesota Driven to Discover
Universities MUniversities Wordmark
Center for Transportation Studies Heading

Webinar Available

A Causal Model of Traffic Conflicts and Crashes: Minnesota’s SHRP2 SO1 Project

Gary Davis, Professor, Department of Civil Engineering, University of Minnesota-Twin Cities

September 23, 2010

At the September 23 Advanced Transportation Technologies seminar, Gary Davis, professor of civil engineering at the University of Minnesota, described the development of analysis methods for data that are currently being collected by the Strategic Highway Research Program 2 (SHRP2).

SHRP2, funded by Congress in 2006 to investigate the underlying causes of crashes and congestion, focuses on four areas: improved safety through the understanding of driver behavior, infrastructure renewal, reduction of congestion and travel time, and planning for new capacity. Davis noted that SHRP2 is unique because it treats safety as a separate focus rather than a component of another activity.

To better understand driver behavior, SHRP2 researchers will conduct a naturalistic study of driver behavior by outfitting nearly 2,000 vehicles in six locations across the United States with an instrumentation package to monitor drivers’ actions. The instrumentation packages include four cameras and front radar. Once the naturalistic driving study is concluded, the collected data will be made available to researchers with the hope that they will provide more real-world information about driver behavior than researchers have ever had, Davis said.

Davis and his colleagues were involved in one component of SHRP2: the development of analysis methods for the data that will be collected in the naturalistic study. This is important because the number of actual crashes among the 2,000 drivers participating in the field study will probably be fairly small. But researchers are interested in events that explain something about crashes even when no crash occurs. To create a model for analyzing the SHRP2 data, Davis and his team needed obtain available data like the data that will be collected in the naturalistic study.

Davis and his team used vehicle-based data from a completed 100-car pilot study, along with site-based data from the Minnesota Traffic Observatory (MTO) and the Cooperative Intersection Collision Avoidance Systems (CICAS) project. Their analysis focused on the statistical relationship between near-crashes, or conflicts, and actual crashes.

Near-crashes have two features: a counterfactual component and an extremity component. In other words, a crash did not occur (“counterfactual”), but there would have been a crash without “extreme” evasive action. The research team’s goal was to develop a method for quantifying these features in actual near-crash events—that is, to express mathematically the degree to which a near-crash could have been a crash.

Davis and his team first created a simple rear-end collision model, using video data collected by the MTO’s advanced detection and surveillance stations, which are located on high-rise buildings overlooking I-94 in downtown Minneapolis. A crash occurs when the distance required for the following car to come to a stop is greater than the available stopping distance. When numerical values are assigned to variables in this model, a mathematical computation can determine whether these values will result in a crash.

Using data from all three sources, the researchers also considered near-crash situations with more complicated trajectories than simple rear-end collisions. For each situation, they described driver behavior as a sequence of accelerations. Using that sequence, along with the initial speed and position of the vehicle, they were able to create and solve an equation to determine the degree to which a near-crash could have been a crash.

Davis and his team came to the following conclusions:

  • Trajectory-based reconstruction is feasible using both vehicle-based and site-based data.
  • It is possible to extend methods of counterfactual analysis to differential equation models.
  • There is limited evidence that the distributions of evasive actions for crashes and near-crashes overlap.
  • The CICAS system, with modifications, could support site-based field research at lower-volume intersections.
  • The usefulness of vehicle-based field data will strongly depend on the ability to calibrate and maintain the data-collection systems.