Research On Key Technologies Of Automatic Parking System Based On Around View Monitor

With the rapid development of intelligent driving technology,automatic parking technology has become one of the hotspots.For automatic parking system,how to accurately detect the parking space on the ground with marking line is a key problem.But in fact,under adverse conditions,such as lack of light,diversity of ground materials and parking space covered by shade,most of the parking recognition methods based on machine vision can not effectively detect the target parking space.At the same time,after completing the parking space detection,how to make the vehicle enter the parking space accurately is another key issue.In order to solve the above problems,this paper studies the two key technologies of parking space recognition and path planning in the automatic parking system based on the gaze system.First of all,the distortion correction,inverse perspective transformation and coordinate system transformation are carried out on the original image installed in the fisheye camera around the vehicle to construct the panoramic image around the vehicle.At the same time,the ultrasonic sensor data analysis and installation,complete the automatic parking perception system.Secondly,a parking identification method based on the loop-view system is proposed.Firstly,a parking data set is established,including 3000 panoramic images of typical parking spaces under different illumination and ground conditions,and the corner points in each image are classified and labeled.Taking the panoramic image around the vehicle collected by the panoramic system as the input,the corner detection algorithm model based on the trained YOLOv4 can accurately detect the region of interest of the corner of the parking space.The image processing operations such as Hough change in the region can obtain accurate corner coordinates of the parking space.According to the corner type of parking space and the distance between adjacent corner points,the type of parking space is determined.Finally,the fusion of ultrasonic sensor distance information can determine whether the target parking space is empty.Then,the vehicle kinematics model is established to study the path planning of parallel and oblique parking spaces.Aiming at the difficulty of parallel parking in narrow space,a parallel parking path planning method with attitude adjustment in the library is proposed.At the same time,the types of oblique parking spaces and parking obstacle avoidance are analyzed.Finally,the path planning of the above two parking spaces is simulated and analyzed.Finally,the hardware and software system of automatic parking system is built on the laboratory intelligent electric vehicle platform.The proposed image stitching,parking recognition algorithm and path planning are verified by the real vehicle,and the accuracy of the image stitching effect,as well as the accuracy,robustness and superiority of the parking recognition algorithm are verified.Finally,the real vehicle experiments on parallel and oblique parking spaces verify the effectiveness of the path planning method and the feasibility of the automatic parking system designed in this paper.