Lane Change Intention Recognition Based on Natural Driving

To anticipate driving intentions in advance, the natural driving trajectory is utilized for predicting driver lane change intentions based on driving style analysis. Initially, Tracker is employed to extract natural driving trajectories from the expressway merging area. Subsequently, K-means++, Spectral Clustering, and DBSCAN are introduced to determine the clustering model that exhibits the best performance in analyzing driving styles. Lastly, the XGBoost, LightGBM, and stacking models are proposed to identify driver lane change intentions and analyze the impact of driving style information on the model. The results indicate that the best outcomes are achieved when K-means++ is used to categorize driving styles into conservative, normal, and aggressive categories. The stacking model exhibits the highest accuracy, while LightGBM demonstrates the fastest training speed. Considering driving style information enhances the model’s performance, yielding an accuracy of over 85% within 1.5 s prior to the lane change initiation.