果园监控机器人自主行驶及视觉导航系统研究  

Research on Autonomous Driving and Visual Navigation System of Orchard Monitoring Robots

作　　者：金李 高何璇 高晓阳 李红岭 杨梅 贾尚云 张旭 唐渲运 JIN Li;GAO HE-xuan;GAO Xiao-yang;LI Hong-ling;YANG Mei;JIA Shang-yun;ZHANG Xu;TANG Xuan-yun(College of Mechanical and Electrical Engineering,Gansu Agricultural University,Lanzhou Gansu730070,China;Gansu Key Laboratory of Viticulture & Oenology Engineering,Lanzhou Gansu 730070,China;Gansu Provincial Key Laboratory of Aridland Crop Science,Lanzhou Gansu 730070,China;Network Finance Department of Lanzhou Bank,Lanzhou Gansu 730000,China)

机构地区：[1]甘肃农业大学机电工程学院,甘肃兰州730070 [2]甘肃省葡萄与葡萄酒工程学重点实验室,甘肃兰州730070 [3]甘肃省干旱生境作物学重点实验室,甘肃兰州730070 [4]兰州银行网络金融部,甘肃兰州730000

出　　处：《林业机械与木工设备》2022年第7期24-31,共8页Forestry Machinery & Woodworking Equipment

基　　金：国家自然科学基金(61661003);学科建设基金(GAU-XKJS-2018-190)。

摘　　要：目前果园的人工巡检在面对大规模区域种植时,作业任务极为繁重。基于此研究了一种在果园道路自主行驶的监控车以及机器视觉导航系统。选用基于图优化的Cartographer算法进行果园路面信息采集以及全局路径规划,采用LiDAR(Light Detection and Ranging)进行道路建图和路径规划,利用机器视觉系统进行道路分割识别并提取道路虚拟中线为导航线,并在转角区域等利用激光雷达定位和路径规划。提出了聚类算法和HSV颜色空间分割算法相结合的道路区域图像分割识别算法,并采取形态学滤波和连通域处理降噪方法,实现了果园道路的准确提取。在视觉导航实验研究中,分块并标记道路目标区域中心点,曲率拟合获得导航线,实验结果表明拟合提取的虚拟导航线与道路实际中线偏差小于4%,准确性高,满足机器视觉导航的精度要求。At present,the manual inspection of orchards involves extremely heavy workloadin the face of large-scale regional planting.Based on this,a monitoring vehicle and a machine vision navigation system for autonomous driving on orchard roads was studied.The Cartographer algorithm based on graph optimization was used for orchard pavement information collection and global path planning.LiDAR(Light Detection and Ranging)was used for road mapping and path planning.Machine vision system was used for road segmentation and recognition,and the virtual center line of the road was extracted as the navigation line,and lidar positioning and path planning were used in the corner area.In this paper,a road area image segmentation and recognition algorithm combining clustering algorithm and HSV color space segmentation algorithm was proposed,and morphological filtering and connected domain processing noise reduction methods were adopted to achieve accurate extraction of orchard roads.In the experimental study of visual navigation,the center point of the road target area was divided into blocks and marked,and the navigation line was obtained by curvature fitting.The experimental results showed that the deviation between the virtual navigation line extracted by fitting and the actual center line of the road was less than 4%,and the accuracy was high,satisfying the accuracy requirements of machine vision navigation.

关 键 词：机器视觉 机器人 图像处理 分割 激光雷达 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]