Research On Control Algorithm Of Vehicle Lane Keeping System

The vehicle Lane Keeping System can improve active driving safety and reduce the driver's burden,which consists of two parts: active intervention and lane departure warning.For the active intervention algorithm,Model Predictive Control(MPC)which is studied widely is computationally intensive,Single-point Preview control is based on software simulation,so there are some distance from the practical application.For the early-warning decision-making algorithms,the current lane departure warning is mainly based on the turn signal and the relative position of the vehicle and the road which is sensed by the camera,so the false alarm rate is high,and when the driver doesn't switch on the turn signal or on the emergency turn,the system may produce false alarm and interfere with the driver's normal driving.In order to solve the above problems,based on the simulation experiment and theoretical analysis,the Single-points Preview driver model is modified to achieve a smooth and error-free tracking of a given path in this work;In the aspect of early-warning decision-making algorithms,the Gaussian Mixture Hidden Markov Model(G-HMM)is used to detect driving intention,at the same time the lane deflection is evaluated by longitudinal TLC(Time to Lane Cross),then the decision is made based on the driving intention and the degree of vehicle deviation.At first,this thesis studies the lane keeping active intervention algorithm: Single-point Preview driver model.The input variables and some of the intermediate variables are defined as measurable errors on the base of the structure of the original Single-point Preview driver model,at the same time,the lateral position error feedback coefficient K is added,then we found the influence of preview time and feedback coefficient on the control effect after the simulation experiment and theoretical analysis.Further analysis shows that the fundamental reason for the existence of steady-state error is that there is no distinction between longitudinal preview time and lateral acceleration time.Therefore,a time coefficient F that characterizes the relationship between lateral acceleration time and the longitudinal preview time is set,after theoretical analysis and simulation,it is proved that the improved model can realize the tracking of zero steady-state error for a given path,and the damping and vibration frequency of the system are adjustable.Then,the driving intention recognition algorithm is designed and validated.The Hidden Markov theory is applied to build a G-HMM which describes driving intentions through steering wheel angle,vehicle lateral position error,longitudinal speed,road curve radius and vehicle yaw angle.The data was collected by the Driver-In-the-Loop experiment,the model parameters were obtained through training.By real-time assessment of the matching of models and observations corresponding to different working conditions,the corresponding working condition of the model with the highest matching degree is the driving intention.The effectiveness of the algorithm is verified through off-line simulation experiments.Finally,a complete lane keeping system control algorithm is built and verified through simulation experiment.The lane departure judgment algorithm is designed based on longitudinal TLC,the lane departure warning and intervene decision was made combined with lane departure judgment and driving intention recognition under certain rules.The virtual driver model under different states is designed and simulated,the results show that the improved Single-point Preview driver model has good effect on the steering wheel angle controlling,the driving intention recognition algorithm can identify the driver's intention accurately,the departure warning and intervention decision algorithm can generate warning signal and make decision normally.