베이지안 확률 추론 기반의 자율주행차 사고 계층분석 및 사고 심각도 모형 연구

초록

With the advancement of Advanced Driver Assistance System technology, autonomous driving technology is get a attention as a key component of smart transportation systems. However, verification of safety about autonomous vehicle is required due to the continuous traffic crash involving autonomous vehicles. In particular, to ensure the safety of vulnerable road users such as pedestrians as well as autonomous vehicle on handicap sections where autonomous mode is unexpectedly disengaged and the driver's response time is insufficient, this requires a technical/institutional approach and a continuous understanding for autonomous vehicle crashes. Therefore, this study used autonomous vehicle crash record data to extract impact factors from the autonomous vehicle collision reports provided by the California Department of Motor Vehicles, and used latent class analysis and ordered probit model with Bayesian probability inference to deal with the traffic crash sample. As a result, traffic crash involving vulnerable road users such as pedestrians and cyclists, scooter, mainly occur at bus stops or bicycle lanes where road users cross, and the traffic crash severity increases when light condition is dark street. On the other hand, for the traffic crash which is not involve vulnerable road users, the crash classes were divided and the traffic crash severity increased depending on weather conditions, autonomous vehicle movement, day and night, etc. When considering unobserved heterogeneity in geometry and weather conditions, it was found that the the variance size was smaller and the confidence interval was narrower based on the Bayesian ordered probit model with random effects unlike fixed effects. Specifically, the traffic crash severity increased mainly in movements including turns around signalized intersections and for the other crash, which does not include the vulnerable road users. This means that intersections require an exploring of vehicle movements due to the cross to the various road users, complexity of traffic flow, and unpredictability of vehicles and pedestrians. By developing a detailed autonomous vehicle safety evaluation scenario on these study, it is expected that more efficient and reasonable safety evaluation of level 4 to level 5 autonomous driving technology will be possible.

키워드