Research On Vision Based Road Surface Recognition Technology And Its Application To Semi-active Suspension Vehicles

The semi-active suspension system with adjustable stiffness and damping has the characteristics of low energy consumption and good shock absorption effect,making it the mainstream direction of vehicle suspension development.Introducing road surface information into the existing semi-active suspension control system can further improve vehicle performance,which is one of the development directions of semi-active suspension.Therefore,this article takes the semi-active suspension system as the research object,and studies visual road surface recognition technology based on monocular camera images.On this basis,further research is conducted on semi-active suspension control technology.This way,when the vehicle is driving on the road,the system can collect road surface images through monocular cameras,and use image recognition technology to obtain road surface models.The controller can based on road conditions The vehicle dynamics model and its motion parameters are used to control the semi-active suspension actuator.To test the effectiveness of this study,a simulation model of a semi-active suspension vehicle system was built using MATLAB/Simulink software,and various typical road conditions of constant speed straight driving and complex road conditions of non emergency braking were simulated.The simulation results showed that the road recognition technology and suspension control technology proposed in this thesis can improve vehicle performance.The specific research content is as follows:On the basis of studying various classic convolutional neural networks and YOLO series algorithms,an improved Darknet-19 network and an improved YOLOV3 tiny algorithm were proposed for road surface features.Subsequently,a Darknet-19 road classification model and a YOLOV3 tiny typical road recognition model were constructed,and the above two models were trained through experiments.Finally,the effectiveness of the constructed model was verified.In order to reduce the difficulty of modeling and generate usability that satisfies the model,equivalent controllable spring force and adjustable drag force are used to describe the semi-active suspension model.Then,based on the characteristics of uneven and block roads in visual recognition,they are constructed into typical and complex road models,which are used as excitation sources for vehicle vertical motion,The semi-active suspension vehicle system dynamics model under non emergency braking condition and the semi-active suspension vehicle dynamics model under uniform straight driving condition are established in turn,and the comprehensive evaluation index of semi-active suspension performance is constructed.In order to improve the performance of vehicles traveling on multiple typical road surfaces at a uniform speed in a straight line,a semi active suspension LQG fuzzy control strategy for multiple typical road surfaces is proposed.The fuzzy control is based on visual road recognition results to suppress vehicle pitch motion,while the LQG optimal control focuses on optimizing the overall performance of the vehicle.Finally,a simulation model was built using MATLAB/Simulink software and simulation experiments were conducted.The simulation results show that when the asphalt road faces the cement road,the root mean square values of the body pitch angle,pitch angular acceleration and vertical acceleration of the LQG fuzzy control suspension are reduced by 63.24%,55.92% and 57.88% respectively compared with the uncontrolled suspension,while the root mean square values of the front wheel dynamic load,rear wheel dynamic load,front suspension dynamic deflection and rear suspension dynamic deflection are reduced by 15.01%,8.95%,50.85% and 44.41%respectively.In order to improve the vehicle performance under non emergency braking conditions on complex roads,a semi-active suspension with suboptimal fuzzy control strategy for complex roads was proposed.The simulation results show that compared with the uncontrolled suspension,the mean square root values of body pitching angle,pitching angular acceleration and vertical acceleration are reduced by 60.64%,61.97% and 55.34%respectively,while the mean square root values of front wheel dynamic load,rear wheel dynamic load,front suspension dynamic deflection and rear suspension dynamic deflection are reduced by 36.74%,41.06%,40.84% and 54.56% respectively.