摘要
随着当代科技日新月异,智能机器人的自主能力快速发展,而其发展面对的挑战之一便是障碍物识别技术的提升。提出一种基于机器视觉的障碍物识别系统,目的为实现快速识别障碍物。本系统首先利用张氏平面标定法得到双目摄像头的内外参数,并将拍摄的图像灰度化且进行滤波处理以及图像锐化处理,并通过立体匹配得到双目视差图,最终利用双目视差与深度的关系,得到当前场景的深度信息,最后根据实时测距的要求,采用YOLO目标识别算法,通过当前场景图片数据集进行训练,对常见障碍物进行识别。
With the rapid development of modern science and technology,the autonomous ability of intelligent robots is de-veloping rapidly,and one of the challenges facing its development is the improvement of obstacle recognition technology.In this paper,an obstacle recognition system based on machine vision is proposed to realize fast obstacle recognition.This sys-tem first,Zhang plane calibration method are used to get the inner and outer parameters of binocular cameras and image of the gray level change and filtering processing and image sharpening processing,and binocular parallax figure is obtained by stereo matching,finally using binocular parallax and depth,the relationship between the depth of the current scene informa-tion,finally according to the requirements of real-time ranging use YOLO target recognition algorithm,Through the current scene image data set for training,common obstacles were identified.
作者
韩成浩
曾繁歌
HAN Chenghao;ZENG Fange(School of Electrical and Computer,Jilin Jianzhu University,Changchun,130022)
出处
《长春理工大学学报(自然科学版)》
2023年第1期88-93,共6页
Journal of Changchun University of Science and Technology(Natural Science Edition)
基金
吉林省发展和改革委员会产业技术研究与开发项目(2020C019)
吉林省科技发展计划项目(20190303096SF)
长春市科技计划项目(21ZY46)。
关键词
机器视觉
自主避障
图像匹配
深度信息
machine vision
autonomous obstacle avoidance
image matching
depth information