What are the Differences in Driver Lane-Changing Intention Models Recognition Performance Between Connected and Non-Connected Environments

In the connected environment, traffic information sources become more heterogeneous, transforming vehicle-road information from static isolation to dynamic interconnection. Therefore, this work aims to examine the effects of the connected environment on Lane-Changing Intention (LCI). Firstly, a driving simulator experiment was designed to imitate the connected environment. Secondly, the LCI time window and feature parameters, including eye movement parameters, vehicle motion parameters, and driver operation parameters, were compared in connected and non-connected environments. The results demonstrated that the LCI time window was longer in the connected environment (6.6 s) than the non-connected environment (4.1 s). Moreover, the LCI feature parameters were significantly different between both environments. Finally, a novel driver LCI recognition model was developed for both environments. This involves utilizing phase space reconstruction and recurrence plot technology to convert time-series feature parameters into images. The swin transformer algorithm was then introduced to classify these images into lane-changing and lane-keeping. Comparing the LCI models in both environments showed no significant difference in model accuracy at 0.5 s before lane-changing maneuver. Interestingly, the model accuracy in the connected environment significantly outperforms that in the non-connected environment during the 2–4 s before lane-changing maneuver. In addition, the novel LCI model accuracy was 90.80% at 3 s before lane-changing maneuver, surpassing that of the traditional machine learning algorithm. To sum up, this research contributes to enhancing the accurate response of lane-changing assistance systems as well as the transfer control right between human and machine co-driving in the connected environment.