Risk Status Identification During the Takeover of Conditionally Automated Vehicles

In order to identify the risk status during the takeover from automated driving to manual driving in dangerous traffic situations, this study investigated their performance during the takeover based on PreScan. The experiment invited 38 drivers. They all held a valid driving license. A risk status identification model was proposed based on the information of vehicle status and traffic environment status. According to the rate of electrocardiogram(ECG) and the performance of takeover, the risk status was classified into three levels. Using Pearson correlation coefficient algorithm, seven factors were selected as the feature set. Then, the algorithm of Random Forest (RF) was employed to establish the takeover risk status identification model. The results show that the accuracy of RF is 98.8%, increasing 10.4%, 17.7% and 7.3% compared with Support Vector Machine (SVM), Classification and Regression Tree (CART) and Back Propagation Neural Network, and each risk level has good prediction results. Respectively, the results show that the space headway, longitudinal acceleration and lane departure have a great influence on the risk level, and the space headway has the strongest impact on the degree of danger during the takeover of control in Conditionally Automated Vehicles.