A Preliminary Model of Driver Acceleration Behavior Prior to Real-World Straight Crossing Path Intersection Crashes Using EDRs

There are approximately 4,500 traffic fatalities at intersections each year in the U.S. One method for reducing these crashes is through equipping vehicles with Intersection Advanced Driver Assistance Systems (I-ADAS) that can detect and alert the driver of an impending crash. However, the effectiveness of these systems is expected to be highly dependent on the pre-crash acceleration behavior of drivers prior to these crashes. The first objective of this study was to use pre-crash acceleration data downloaded from Event Data Recorders (EDRs) to evaluate previously developed acceleration profile models that were based off of "typical" acceleration data. Our second objective was to develop a novel acceleration profile model using this pre-crash data. This study used EDR-recorded pre-crash speeds from 89 intersection crashes investigated as part of National Automotive Sampling System / Crashworthiness Data System (NASS/CDS). A 10-fold cross-validation procedure was implemented to generate and evaluate a quadratic-decreasing acceleration profile model developed from this pre-crash data. Our results indicate that previously generated models underestimate pre-crash acceleration behavior of vehicles involved in intersection crashes. An improved, nationally-representative acceleration profile is also presented that should be considered by researchers interested in simulating pre-intersection crash acceleration behavior. These findings have important implications for I-ADAS designers and government regulatory agencies developing and evaluating the algorithm for delivering an effective, reliable alert to the driver.