巡检机器人导航系统研究与设计  

Research and design of navigation system for patrol inspection robot

作　　者：宋明磊 谢丽蓉[1] 安冬 SONG Minglei;XIE Lirong;AN Dong(School of Electrical Engineering,Xinjiang University,Urumqi 830047,China;College of Information and Electrical Engineering,China Agricultural University,Beijing 100083,China)

机构地区：[1]新疆大学电气工程学院,新疆乌鲁木齐830047 [2]中国农业大学信息与电子工程学院,北京100083

出　　处：《现代电子技术》2022年第24期23-29,共7页Modern Electronics Technique

基　　金：国家重点研发计划(2019YFD0901004⁃01)。

摘　　要：针对渔场巡检导航出现的行驶速度慢、安全系数低等问题,文中提出一种RTK⁃GAN⁃CNN⁃MFAC组合导航系统。在大范围内,使用实时差分定位(RTK)获取巡检机器人的准确位置,通过无模型自适应控制(MFAC)方式控制巡检机器人运行速度并按照既定路线行驶。采用单目相机获取障碍物图片数据,通过生成式对抗网络(GAN)对图片进行处理;然后设计卷积神经网络(CNN)训练分类模型,通过模型对巡检机器人行驶中的实时图片进行分类处理,若识别到巡检机器人前方存在障碍物,则切换为GAN⁃CNN⁃MFAC视觉避障,对障碍物进行避障。实验结果表明,RTK⁃GAN⁃CNN⁃MFAC组合导航系统在无障碍物时运动速度较快,遇到障碍物时可以安全避开,具有良好的有效性和实时性。In allusion to the problems of low driving speed and low safety factor in fishing ground patrol navigation,an RTK⁃GAN⁃CNN⁃MFAC integrated navigation system is proposed.In a large range,the real⁃time kinematics(RTK)is used to obtain the accurate position of the patrol inspection robot,and the model free adaptive control(MFAC)mode is used to control the speed of the patrol inspection robot and make it drive along the established route.The monocular camera is used to obtain the image data of obstacles,and the images are processed by means of the generative adversarial networks(GAN).The training classification model of convolutional neural networks(CNN)is designed to classify and process the real⁃time images of the patrol inspection robot during driving.If an obstacle is identified in front of the patrol inspection robot,it will switch to GAN⁃CNN⁃MFAC visual obstacle avoidance to avoid the obstacle.The experimental results show that the RTK⁃GAN⁃CNN⁃MFAC integrated navigation system can move faster when there is no obstacle,and can avoid obstacles safely.It has good effectiveness and real⁃time performance.

关 键 词：组合导航 巡检机器人 位置获取 运行速度控制 图片数据获取 图片处理 分类模型设计 视觉避障 

分 类 号：TN820.4-34[电子电信—信息与通信工程] TP249[自动化与计算机技术—检测技术与自动化装置]