Research On Intelligent Vehicle Lane Keeping Technology Based On Information Fusion

With the progress of automobile industry and artificial intelligence technology,autonomous driving technology has been greatly developed.Lane keeping system,as one of the core technologies of autonomous driving,has been widely studied in recent years.In this paper,the failure of vision sensor and the adaptability of controller in lane keeping system are studied.Firstly,the intelligent vehicle system and the bottom control system are introduced in detail,including the architecture of the intelligent vehicle,the principle of sensor selection and the improved design of the bottom control system.Then,the bottom control system of steering,braking and driving of the intelligent vehicle is tested,which shows that the modified bottom control system can meet the requirements of real-time and control precision of automatic driving.Secondly,through the analysis of the actual road test of the vision sensor,it is found that the detection failure of the vision sensor is caused by the absence or occlusion of the lane marking.To solve this problem,a compensation strategy for lane is proposed,which uses sensor fusion to obtain the vehicle trajectory and predicts the lane coefficient based on the relative position relationship between the vehicle and the road.Through simulation verification of the compensation strategy,it is concluded that the road compensation strategy can ensure the accurate prediction of lane coefficient within the time of sensor information loss of 1-2 seconds.Thirdly,linear time-varying model predictive control method is adopted to track the lane in lane keeping system.In order to fully consider the relationship between longitudinal and lateral of vehicle,the vehicle control prediction model is derived based on the 3-dof vehicle dynamics model and the magic formula tire model,and the nonlinear prediction model is locally linearized.In order to improve the vehicle control stability,the control quantity,control increment,output and side angle constraints are introduced,and the constrained objective function is transformed into a standard quadratic programming problem for solving.The simulation platform is built to verify the robustness and adaptability of the designed linear time-varying model predictive controller to road adhesion,vehicle speed change and reference trajectory.Finally,the lane compensation strategy and lane keeping linear time-varying model predictive controller are verified in real vehicles.The results show that the lane compensation strategy improves the stability of the vision sensor to a certain extent and effectively avoids the sudden change of vehicle control caused by sensor failure.At the same time,the lane keeping controller performs well in the track tracking process,and the computing speed meets the real-time requirements.