摘要
针对智能停车库中的泊车机器人视觉系统研究需求,提出一种基于双目视觉的泊车机器人障碍物识别系统。通过双目摄像头进行图像采集,利用张正友棋盘标定法进行双目相机标定;采用Bouguet进行立体校正,将高斯滤波与拉普拉斯算子相结合进行图像预处理;采用YOLO卷积神经网络对目标障碍物进行快速识别;利用区域匹配算法进行立体匹配并生成目标障碍物视差图;通过成像点和目标障碍物的立体几何关系计算得到目标障碍物的深度信息。实验结果表明,该系统具有良好的实时性和较高精度,障碍物识别时间平均为0.0901s,在2600mm具有最佳测距精度,可为泊车机器人自动泊车提供保障。
Aiming at the demand of vision system in the research field of parking robot and intelligent parking garage,a parking robot obstacle recognition system based on binocular vision is proposed.This system collects images with a binocular camera,uses the Zhang Zhengyou checkerboard calibration method for binocular camera calibration;Use Bouguet for stereo correction.Gaussian Filtering and Laplacian are used for image preprocessing;Use YOLO convolutional neural network to quickly identify target obstacles.The region matching algorithm is used for stereo matching and the target obstacle parallax map is generated.The depth information of the target ob?stacle is calculated by the three-dimensional geometric relationship between the imaging point and the target obstacle.The experimen?tal results show that the system has good real-time performance and high precision.The obstacle recognition efficiency is 0.0901s on average,and it has the best ranging accuracy at 2600mm,which can guarantee the automatic parking of parking robots.
作者
王帅
杨建玺
WANG Shuai;YANG Jian-xi(School of Mechanical and Electrical Engineering,Henan University of Science and Technology,Luoyang 471000,China)
出处
《软件导刊》
2019年第12期26-29,共4页
Software Guide
基金
河南省重点攻关项目(1621002210048)
河南省教育厅自然科学研究项目(2010B460010)
关键词
泊车机器人
双目视觉
相机标定
立体匹配
YOLO卷积神经网络
parking robot
binocular vision
camera calibration
stereo matching
YOLO convolutional neural network
