Automatic Parking System Research Based On Machine Vision

With the rapid development of driverless technology,automatic parking technology has become an indispensable part of intelligent vehicles.The principle of ultrasonic radar ranging is used to identify the target parking space by most of the existing automatic parking assist systems,which often leads to the failure of automatic parking,the recognition accuracy is not high,and they can not identify the scene with no obstacles on both sides of the target parking space.In this paper,the automatic parking system based on machine vision is studied by using the fish eye lens.Firstly,based on machine learning algorithm,a vehicle position recognition method which only relies on visual information is proposed.Harr-like feature and Ada Boost algorithm are used to train the vehicle position corner classifier,and a series of operations are carried out on the recognized vehicle position corner area to extract the accurate information of the vehicle position corner.According to the distance between two adjacent corner points,the vehicle position area can be extracted,so a method of parking space state judgment based on similarity cosine is proposed.The gray histogram of parking space area is converted into vector mode and compared with the vector mode of gray histogram of the front sub area of parking space.Their cosine similarity is calculated to determine whether the target parking space is empty or not.Secondly,the path planning of automatic parking is analyzed.According to different parking scenes,the feasibility and characteristics of different parking tracks are analyzed.Finally,a passenger car is used as the test platform to verify the proposed parking space recognition method,and the automatic parking experiment is carried out.The communication between the vehicle and the computer is completed by using CAN-BUS,and the control of shifting,acceleration,braking and turning of the vehicle is realized by sending corresponding messages.The horizontal parking and vertical parking experiments are carried out by using the rear wheel speed sensor of the vehicle for positioning.The experimental results show that the parking space recognition method proposed in the paper has high accuracy,high parking success rate and high system reliability,which verifies the effectiveness and accuracy of parking space recognition and decision-making planning strategies.