Real-Time Risk Identification and Prediction for the Target Lane’s Following Vehicle during Lane Change

Abstract

Real-time risk identification and prediction can assist a driver’s decision-making to ensure driving safety during lane change (LC). Previous studies seldom focused on the risk for the target lane’s following vehicle (TFV). In addition, temporal risk and spatial risk are often ignored in risk quantification. Key features in different LC scenarios deserve more discussion, as well. To address these gaps, this study proposes a framework, including: a real-time temporal and spatial risk quantification indicator for TFVs, risk labeling by the clustering method, and risk prediction and key features comparison by machine learning. LC events and candidate features were extracted from the Shanghai Naturalistic Driving Study (in China), and all events were divided into three scenarios based on road types: freeway, expressway, and surface road. K-means was adopted to classify risk into three levels: high-risk, medium-risk, and low-risk. The extreme gradient boosting (XGBoost) algorithm was applied to predict the real-time risk. Key features comparison results emphasized the importance of interaction features between TFVs and surrounding cars in all scenarios, especially for the relative acceleration. In the freeway scenario, the most important interaction object was the lane-changing vehicle, while for the other two scenarios it was the leading vehicle in the target lane. Relative distance became a more important factor for risk in the expressway and surface road scenarios, and the importance of the relative speed increased significantly in the surface road scenario. These findings can guide the design of advanced driving assistance systems for TFV during LC.