Institute of Transportation, MOTC

        Recently, there have been many crashes, which have caused significant social costs to our country. The safety management of Transportation industry seems to be an important issue. Investigating the factors behind traffic safety-related events is key to reducing the frequency and severity of crashes. Natural Driving Data (NDS) presents the complete process of the crashes. However, there is a lack of consensus on how to define safety-related events, which vary with purpose, circumstances, and available data. The purpose of this study is to develop an effective method for framing safety-related events. We systematically screen crash surrogates and find key influencing factors. This study uses the driving dynamic data and video data of national highway passenger transportation and selects 913 events. We have comprehensive driving information through the data, including environment, driving dynamics, and driver response. First, we use the ADAS to filter out target events with longitudinal driving risk and observe the detailed process from the whole event. Next, we classify the driving dynamics and risk levels, and then use the ROC curve method to establish the best indicators, thresholds, exposure time, and analysis unit for the crash surrogate. Finally, the ROC curve regression model is used to analyze the key factors for identifying the crash surrogate. The research results point out that the covariates that affect the identification of crash surrogates are the driving dynamics of the vehicle and the leading vehicle. Besides, the minimum distance during the event and the average speed are important factors affecting crash surrogates. The indicators TETG and TERCRI are the most suitable indicators, AUC can reach 80%, and the threshold values are between 1.5 seconds to 2 seconds and 6 seconds to 7.5 seconds. In future applications, it can be used as a reference for driver safety management.