Research On Obstacle Avoidance Control Of Intelligent Vehicles Based On Visual Perception Information

With the continuous development of intelligent and electric vehicle technology,autonomous driving has become a new trend in the industry.Autonomous vehicles,also known as intelligent vehicles,use onboard sensors to perceive the surrounding environment of the vehicle and control the vehicle to safely and reliably travel on the road based on the obtained road environment information.The semantic segmentation technology of the free space has important research value in the field of environmental perception of autonomous vehicles,but existing visual detection methods are difficult to meet the requirements of detection accuracy and speed for autonomous vehicles.Therefore,visual detection technology still faces significant challenges in vehicle applications.In addition,how to let the vehicle autonomously and safely complete control actions such as avoidance and lane change under the premise of knowing the vehicle’s driving environment information is also one of the urgent problems for the vehicle autonomous control system.In view of the above problems,this thesis conducts research on road obstacle detection,vehicle obstacle avoidance control,and joint simulation,and conducts obstacle avoidance simulation experiments.Firstly,configure the simulation software and vehicle model used in the vehicle obstacle avoidance control model;The tire model of the vehicle model is introduced,and the Vehicle dynamics model is established based on the idealized assumption of the vehicle system,which is used for the design of vehicle obstacle avoidance control.Secondly,in response to the multi-scale problem of objects in image segmentation,combined with the characteristics of attention mechanism itself,a deep attention joint road obstacle segmentation method was studied.By utilizing the feature attributes between each convolutional layer of the neural network,the lightweight DW convolution and attention mechanism were combined to better fuse multi-scale information while ensuring the overall lightness of the model.In addition,in order to prevent the network from overfitting,the random deactivation strategy of convolutional kernel is introduced.The network model was tested on the Cityscapes dataset and compared with similar methods.The results showed that the studied method had significant improvements in detection accuracy and other aspects.Then,based on navigation or detection methods to locate lane information,a reference path is obtained,and the optimal obstacle avoidance path is planned based on a quintic polynomial function.Decouple the vehicle horizontally and vertically and control them separately.Use LQR as the vehicle’s lateral obstacle avoidance control strategy,and control the vehicle’s lateral displacement and yaw angle by controlling the front wheel angle;And a tracking error model was established to design the steering system,and feedforward control was introduced to eliminate feedforward errors.Under the condition of double line shifting,the lateral controller was verified to be able to stably track the lateral trajectory using Simulink/Carsim software.A neural network PID vehicle longitudinal obstacle avoidance controller was designed by optimizing PID parameters using neural networks.Under NEDC operating conditions,the vehicle longitudinal controller has been verified to have stable and accurate tracking control effects.Finally,the lateral and longitudinal vehicle controllers and path planning algorithms for autonomous vehicles were built in Simulink,Carsim,and Pre Scan simulation software,and urban road scenes were simulated.Joint simulation of the vehicle’s lateral and longitudinal controllers was conducted using perceptual information,and obstacle avoidance control experiments were conducted on different front vehicle motion states.The results indicate that the controller designed in this thesis can effectively complete vehicle obstacle avoidance and lane changing operations after considering the road environment.