Driving Behavior Characterization in Expressway Merging Areas Based on Roadside LiDAR

There is a strong correlation between driving behaviors and traffic conflicts, particularly in expressway merging areas where there are numerous interchanges between multiple lanes. High-resolution vehicle trajectory data are necessary for the analysis of driving behaviors, but traditional information gathering platforms or techniques struggle to meet the practical requirements in terms of usability and portability. To the end, this article builds a roadside LiDAR platform and constructs a high-precision vehicle trajectory database for driving behavior characterization. Based on the constructed database, ten indicators are selected as key features for driving behavior analysis and four classification algorithms including random forest (RF), support vector machine (SVM), backpropagation (BP) neural network, and XGBoost are adopted to categorize driving behaviors into conservative, stable, and radical types. By comparison, the XGBoost achieves the best classification performance among the four algorithms. In addition, the variable analysis is conducted to examine the significance of the indicators on driving behaviors, and it is found that the merging distance, merging speed, relative speed, and relative distance have the greatest influence on merging vehicles to make driving decisions, and these indicators can be considered as inputs of potential conflict assessment models and further enhance the safety and efficiency of the overall transportation system.