重庆理工大学学报2025，Vol.39Issue(21)：46-54,9.DOI:10.3969/j.issn.1674-8425(z).2025.11.006
实车数据驱动的换道意图识别研究
Research on lane-change intention recognition driven by real-world vehicle data
摘要
Abstract
To enhance ADAS'(advanced driver assistance systems)ability of recognizing drivers' lane-change intentions and improve vehicle lane-change safety in dynamic traffic environments,this paper develops lane-change intention recognition models based on random forest(RF)and extreme gradient boosting(XGBoost),and systematically compares their performance under different lane-change time windows.A total of 852 real-world lane-change samplesare collected through road experiments.The datasets are divided into training and testing sets in a 3∶1 ratio and normalized using the min-max preprocessing.During model construction,key hyper-parameters are optimized using grid search.The recognition performance is assessed based on confusion matrix,ROC curve,and F1 score metrics.Experimental results show the RF model consistently outperforms the XGBoost model across 1～5 second lane-change windows.However,as the prediction time window extends to 5 seconds,the recognition performance of both models slightly declines.Further feature importance analysis at 2 and 3 second time windows indicates the steering wheel angle consistently appears as the most critical feature.These findings validate the robustness and adaptability of the RF model in dynamic driving intention recognition tasks and provide both some insights into intelligent driving decision-making systems in complex multi-lane traffic scenarios.关键词
交通系统安全/换道意图识别/随机森林/极端梯度提升/特征重要性Key words
traffic system safety/lane-change intention recognition/random forest/XGBoost/feature im-portance分类
交通工程引用本文复制引用
LYU Jingliang,ZOU Liyang..实车数据驱动的换道意图识别研究[J].重庆理工大学学报,2025,39(21):46-54,9.基金项目
黑龙江省重点研发计划项目(JD22A014) （JD22A014）
