摘要
视觉系统是农业果蔬采摘机器人的重要组成部分之一,很大程度上决定了农业果蔬采摘机器人的采摘速度和质量。本团队基于QtE5.7+opencv3.2+ARM cortexA9开发平台,设计一种基于嵌入式ARM的苹果采摘双目视觉系统。在目标检测方面,提出了一种基于线性融合的苹果目标检测方法;在立体匹配方面提出了一种结合视差范围估计的ADcensus算法,这两种算法运算量小,适合应用在资源有限的嵌入式平台上面。在实验室进行目标检测算法验证和立体匹配算法测试时,该设计能快速提取出苹果目标,结合视差范围估计的ADcensus算法,其相对误差最小为0.96%,用时为5.6 s,能满足果蔬采摘机器人对苹果识别定位提取的速度与精度要求。
Vision system is one of the important components of agricultural fruit and vegetable picking robot,which largely determines the picking speed and quality.Based on QtE5.7+opencv3.2+ARM cortex A9 commercial development platform,an apple picking binocular vision system is designed.In the aspect of target detection,our team propose a linear fusion method,which can detect Apple target quickly.In the aspect of stereo matching,an ADcensus algorithm combined with disparity range estimation is proposed.The algorithm has less computation and high matching accuracy,and is suitable for application on embedded platform with limited resources.Experiments and field tests show that the design can not only quickly extract and recognize Apple targets in complex environments,but also achieve accurate ranging and real-time display of apple targets.Using ADcensus algorithm,the minimum relative error is 0.96%and the time cost is 5.6 s,which meets the speed and accuracy requirements of Apple recognition and location extraction for fruit and vegetable picking robots.
作者
王磊
李会容
周科
牟波
WANG Lei;LI Hui-rong;ZHOU Ke;MU Bo(Institute of Intelligent Manufacturing,Panzhihua University,Panzhihua 617000,SiChuan,China)
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2020年第1期71-80,共10页
Journal of Optoelectronics·Laser
基金
纳米氧化钒粉末制备新工艺及其在锂离子电池正极材料中的应用(省(部)级重大(重点)(2019YJ0687))资助项目。
关键词
机器视觉
双目测距
立体匹配
嵌入式
目标提取
machine vision
binocular ranging
stereo matching
embedded
target extraction
作者简介
牟波(2000-),男,在校本科生,研究方向为嵌入式和图像处理.E-mail:2805010242@qq.com。
