Research On Vehicle Control Of Self-driving Vehicle Based On Visual Servoing

As an important part of the intelligent transportation system,self-driving vehicle has become one of the important directions of the future development of the automobile industry.The key technologies of which include vehicle environment perception,lane location and navigation,macro and local path planning,vehicle motion control,etc.which becomes a multi discipline and cross application technology combination.The self-driving vehicle can significantly reduce traffic accidents and is expected to achieve zero casualty of road traffic,which is more conducive to improving road capacity,reducing congestion,saving energy and improving living environment.In the key technology of self-driving,after global path obstacles changed,considering the vehicle handling stability and ride comfort,how to quickly and effectively plan the local path for safe driving and control the vehicle to avoid obstacle according to the existing sensor information has been the focus and difficulty.In this paper,on the basis of the existing methods,a new control strategy is established with the real-time images acquired by the visual system to enhance the real time capability and algorithm adaptability from local path planning to the vehicle control considering the vehicle kinematics model and the dynamic constraints.The main research contents include the following aspects:1)Analyze the self-driving vehicle research results both domestically and abroad,determine the key problems that need to be solved and the technical route of this article.2)Structure the visual system for self-driving vehicle,including selection,platform construction and camera calibration.Define the reference coordinate system of the vehicle model and establish the vehicle kinematics model and dynamics constraint model.In order to meet the requirements of the linear time variant model in the model predictive control,the nonlinear system of vehicle is linearized.3)Based on the improved covariance matrix,a particle filtering algorithm with multiple eigenvalues fusion is established for the recognition and tracking of vehicle visual targets.As the target state for vehicle identification is nonlinear,and the noise distribution is non-Gaussian,pre-set a large range of interest recognition areas and integrate visual target features with covariance matrix to improve the accuracy and real-time performance of traditional particle filter algorithm,so that the stable and effective tracking of visual target can be realized.4)A segmentation strategy for important regions of the images is proposed and the local path replanning is realized combined with the tentacle algorithm.Local path replanning is an important way to solve the lane changing,follow-up driving and obstacle avoidance.Based on the real-time images of the visual system,a segmentation strategy for important regionsof the images is creatively proposed to three-dimensionally visualize the front area of the vehicle and complete optimization of the whisker path fusing the vehicle dynamic constraints.5)According to the optimal path of the local path planning,the trajectory is tracked by model predictive control,and the control of the vehicle is realized.To meet the requirements of vehicle operating stability and ride comfort,the selection of the whisker path is constrained with the kinematics and dynamics conditions of the vehicle.A model predictive control system with tentacles trajectory planning is built to achieve horizontal and vertical control of the vehicle combining the longitudinal strategy of the vehicle’s minimum safe distance.Finally,the Panosim/Simulink co-simulation platform is used to verify the feasibility and effectiveness of the proposed control strategy.