Abstract
Research Objective:
This research aims to develop a methodology to identify of the driver performance by using on-board vehicle sensors including the information of driving environment. The identified driver performance level will be applied to the design of advanced driver assistance system with individual adaptation. However, it is important to clarify the features which can identify the driver performance with high accuracy without depending on the direct measurement of the physiological data.
Methodology:
This paper proposes a model-based driver performance level identification method by using the vehicle-control driver models. For the longitudinal direction, the car-following driver model uses the headway distance and the relative velocity to predict the acceleration used for keeping the appropriate distance with the preceding vehicle. For the lateral direction, the 1st order preview-predictive lane-keeping driver model uses the lateral displacement and the yaw angle to predict the steering wheel angle. Using the model identification from the driving data in good condition, the reference driver model with identified parameters can be obtained. Considering a driver as a closed-loop feedback control system, the driver performance level is estimated by comparing the real-time driver operation data and vehicle motion data with those estimated from the reference model. This study is based on the assumption that when the deviation from the reference model becomes large, the driver performance level is degraded. Statistical indices such as the average and the variance of various driving data are used to quantitatively identify the driver performance level.
Results:
From the data analysis, the reference driver models in longitudinal and lateral directions can express the driver behavior in good condition with certain accuracy. To validate the proposed method, the reference driver models are used to identify the driver performance level from the naturalistic highway driving database using the experimental car with a drive recorder, as test data. Low-performance driving data is extracted from the driving data in drowsy state. The drowsiness level is independently graded by the reviewers from the collected driver face images. The experimental results show that using the proposed longitudinal and lateral driver models can estimate the driver performance level which shows the same trend with the independent performance level annotation by the reviewers.
Limitations of this study:
The proposed estimation method assumes the case that there is no road inclination, bank, sharp curves. Such disturbances may affect the accuracy of the model. Digital map information can be taken into account to improve the accuracy of the proposed identification method.
What does the paper offer that is new in the field in comparison to other works of the author:
This paper proposes the method which embeds two driver models inside to estimate the driver performance level by looking at the correlation between the input and the output from the viewpoint of control engineering. This method is superior to the conventional “attention assist system” which only monitors the vehicle motion data.
Conclusion:
This paper presented the theoretical model-based methodology to identify the driver performance level by analyzing the deviation from the reference driver model synthesized by the data collected by the continuous logging drive recorder without using physiological data. The effectiveness of this methodology on estimating the drowsiness level is verified in the paper using the driving simulator and the experimental vehicle.
KEYWORDS – Driver Assistance System, Driver Model, Driver Performance Level, Driver State Estimation, Model Identification