(Completed in Dec/2021)
In order to guide the automatic driving test safety officer (hereinafter referred to as the driver) to correctly control the automatic driving system to deal with various emergencies encountered in daily driving, a system for testing the driver's ability to reflect was developed to realize the test driver based on the monitoring of the reaction ability, the model and classification combination were used to rank all the test drivers through the collected data, reflecting the general reaction ability of each driver (the shorter the reaction time, the higher the ranking).
The test collected the responsiveness of 50 drivers in 3 different stages, including the undisturbed responsiveness test (stage 1), the responsiveness test under manual driving (stage 2), and the responsiveness test in automatic driving mode (stage 3), and obtained a total of 28 features such as reaction time by the driver clicking the mouse and stepping on the pedal to form the test data. The features in the test data were clustered using K-means clustering and Gaussian Mixture Model, and the obtained cluster target variables were predicted and modeled using six classifiers to obtain the feature weights of the testers, and then calculatedthe response ability score.
The research results show that: after comparing the model accuracy and the number of features considered, it is concluded that the combination of Gaussian mixture model clustering and extreme random tree classification has the best performance, and at the same time, almost all attributes can be considered and the allocation is more reasonable. The weights of the three well-performing combined models are combined, and finally it is concluded that compared with the hand, the driver's response ability and the driver's foot response error are significantly higher. And among the indicators of foot response, the driver's response ability and the number of errors in the driver's foot response have the highest significance, indicating that there is a correlation between the two, but the causal relationship between the two needs further research.