Genetic algorithm-based testing scenarios selection for the performance evaluation of crash imminent braking systems for pedestrian safety

Improving driving safety is one of the most important tasks for intelligent transportation systems. In recent years, active safety systems have been widely investigated and have been playing an important role in driving safety. Among these systems, pedestrian crash imminent braking systems are towards pedestrian safety and are able to predict potential crash/near-crash events associated with pedestrians and vehicles, and take appropriate actions to mitigate potential crash effects. In order to evaluate the performance of such systems, we need to design suitable testing scenarios, collect field data on the system performance, and analyze data to see how the systems behave for certain performance indices. Clearly, due to the time and cost of vehicle field testing, it is impossible to test every scenario for performance evaluation. Therefore, a subset of test scenarios, which are critical (i.e., in terms of crashes, fatalities, social cost, etc.) and capture key crash parameters, must be obtained for vehicle testing. In this paper, we propose a methodology that is based on Genetic Algorithm (GA) to find a subset of testing scenarios from a complete set of testing scenarios (obtained via crash databases), by satisfying given constraints. We show that our GA-based approach is effective and efficient. More specifically, it can find the approximated optimal testing scenarios in a much faster time than exhaustive search method. Field data are also used to validate our approach on the testing scenarios selection.