Development of an FCW Algorithm Evaluation Methodology with Evaluation of Three Alert Algorithms

摘要

The report explores the use of real crash data collected in a naturalistic driving study to investigate the potential of collision avoidance systems in avoiding rear-end crashes. In the research effort, three previously proposed collision avoidance algorithms were modeled in software and real crash and near-crash data were input into the algorithm models. The timing of the warnings generated by the algorithms was then evaluated to estimate the percentage of the driving population who would be able to respond to the warning in time to avoid collision based on kinematic estimates of the braking needed to avoid collision. The methodology provides useful guidance both in estimating benefits achieved by the algorithms and in estimating frequencies of alerts in normal driving situations. The method could be improved by accommodating different reaction-time estimates based on whether the driver was already braking at the time he or she detected the event. Estimation of the frequency with which crashes similar to those tested occur across the country, and adjustment of benefit estimates accordingly, may also be informative. The method provides an additional approach for system developers during testing and evaluation.