Prediction of unintended lane departure based on virtual driver

To address the unintended lane departure problem in nowadays´ vehicle safety, a novel method combining the quantitative description of criterions used for evaluating human driving behavior is proposed in this paper. The quantitative analysis of criterions incorporates vehicle kinematics, so can provide “virtual driver” sufficient basis for monitoring the human driving action. Based on that, critical trajectory is proposed to describe possible trajectory domain of the vehicle in short future time, instead of using the measured steering angle. This point is useful to conquer the disturbance of measuring error of the steering error, especially driving at high speed. Another aspect, the method doesn´t need a lane model to describe the center line of the lane, but just uses the detected lane boundary points to predict tendency of lane departure, so can overcome the influence of lane recognition problem. Based on the novel method, the human driving action is graded at three levels: safe, remind, warn. Real vehicle experiments based on the method are carried out on the prototype HONGQI Autonomous Land Vehicle on highway. Results show that the method can predict the tendency of human driving action well, so can give rational suggestion when needed.