Research On Obstacle Recognition System Of Parking Robot Based On Binocular Vision

With the improvement of people's quality of life and the continuous development of the economy,the number of car ownership has increased by spurt,and parking has become a difficult problem in daily travel.Therefore,the intelligent parking garage came into being,and the parking robot as its core vehicle has attracted more and more attention and becomes a new key application in the field of mobile robots.In view of the fact that the current domestic and international parking robots use high-cost laser sensors,the technical difficulty is unique,and it is not easy to realize the problem of large-area promotion.This paper replaces the laser sensor with a low-cost visual sensor that can obtain the surrounding environment information in real time.Identification of obstacles on the parking path.The main research work is as follows.(1)The construction of obstacle visual recognition system.By studying the principle and mathematical model of binocular vision system,combined with the application environment of this system,the software and hardware platform of the system is built.Based on the analysis of the general steps of the binocular vision system,the workflow of the system is designed.(2)Calibration and correction of obstacle recognition system based on binocular vision.On the basis of studying the calibration method of binocular camera,the advantages and disadvantages of the binocular camera are analyzed,and the Zhang Zhengyou chessboard calibration method is finally selected.By comparing and analyzing the advantages and disadvantages of OpencCV calibration method and Matlab calibration method,the Matlab toolbox is used to perform binocular camera calibration,and the control variable method is introduced to carry out experiments and precision analysis on different calibration distances and checkerboard specifications to obtain the optimal precision focal length.Finally,Binocular stereo correction is performed by using the OpenCV-based Bouguet algorithm to obtain parallel and linealigned standard image pairs.(3)Research on stereo matching algorithm in parking garage environment.Firstly,the necessity and existing problems of stereo matching algorithm are introduced.Analyze the matching algorithm and its application environment,and choose he feature-based matching algorithm,For the traditional feature matching algorithms SIFT and SURF,there is a problem that the matching time is long and the feature points of the edge smooth target are weakly extracted.carry out improvement research and design a stereo matching algorithm applied in the parking garage environment.(4)Research on obstacle category detection.There are problems such as less object detection,low accuracy and high time complexity in traditional object detection algorithm.In response to these problems,this paper introduces YOLO convolutional neural network based on deep learning to achieve accurate and real-time detection of obstacle categories.The network has strong scalability,and can train its own sample data set according to actual needs to be applied to different scenarios.The experimental results show that the average error of the obstacle recognition algorithm based on binocular vision in the unmanned intelligent parking garage environment is under 50 mm in the range of 2100 mm,the average time of obstacle category detection is 0.096 s,and the average time of the system is 0.466 s,it meets the real-time and accuracy requirements of the parking robot in the intelligent garage environment for obstacle detection,lay the foundation for the research and application of the parking robot with binocular vision.