Lane-Changing Recognition of Urban Expressway Exit Using Natural Driving Data

The traffic environment at the exit of urban expressway is complex, and vehicle lane-changing behavior occurs frequently, making it prone to conflict congestion. To study conditions improve road operation capacity, this paper analyzes characteristics behaviors exit, adds driving style into influencing factors lane-changing, recognizes one’s intention based on data. A UAV (unmanned aerial vehicle) used collect natural track data diverge area, segments that meet standards are extracted, 374 obtained. K-means++ cluster which grouped three clusters, corresponding “ordinary”, “radical”, “conservative”. Through random forest model identify predict style, accuracy reaches 93%. Considering a single time point historical window, XGBoost, LightGBM, Stacking fusion established recognize intention. results show models can well drivers. has highest accuracy, while LightGBM takes less time; considering window performs better than other one, prediction behavior.