Identifying Driving Risks for Regular Intercity Buses Based on Video Data

Regular intercity buses are the most common type of vehicle among the tourist chartered buses, liner buses and dangerous goods transport vehicles, involving the largest number of passengers and the longest operational cycles. Accidents of them often result in immeasurable loss of life and property damage. This paper zeroes in on the driving scenarios of regular intercity buses, analyzing video data of drivers from the 10 s prior to system alarms. Abnormal driver behaviors are manually identified, and driving risks are categorized into low, medium, and high levels based on the frequency of alarms. The importance of variables is ranked using the Random Forest algorithm, and machine learning algorithms are combined to identify and predict driving risks of regular intercity buses. The study investigates the impact of driver behaviors on risk and provides reference for subsequent risk prevention. Our findings demonstrate that the risk identification effect of regular intercity buses is best when using the XGBoost algorithm combined with the top 12 most important variables, with accuracy 0.8116, precision 0.8218, recall 0.8125, and F1 0.8129. Driver behaviors such as blinking, closing eyes, chatting, and driving with one hand should be closely monitored.