结合YOLOv8与双目测距算法的水面漂浮垃圾检测定位系统设计  

Design of water surface floating garbage detection and positioning system combining YOLOv8 and binocular ranging algorithm

作　　者：何君尧 王文胜 韩宜航 HE Junyao;WANG Wensheng;HAN Yihang(Mechanical and Electrical Engineering School,Beijing Information Science and Technology University,Beijing 100192,China)

机构地区：[1]北京信息科技大学机电工程学院,北京100192

出　　处：《现代电子技术》2024年第20期1-7,共7页Modern Electronics Technique

基　　金：国家重点研发计划课题(2020YFB1713205);北京市教育委员会科研计划项目资助(KM202411232023);2024年北京信息科技大学“青年骨干教师”支持计划(YBT202403)。

摘　　要：为解决户外水域垃圾自动回收船的垃圾目标定位与识别差的问题,提出一种结合YOLOv8与双目测距算法的水面漂浮垃圾回收船的垃圾识别定位系统。该系统主要由摄像头、上位机视觉处理单元和下位机控制单元三部分组成,通过对水域环境内的垃圾进行视觉识别分类后,再进行定位和测距,实现水域垃圾的定位和识别;在定位和识别之后,控制机器收集垃圾。采用双目相机获取图像,使用Jetson Nano嵌入式芯片作为上位机主控芯片,利用最新深度学习模型YOLOv8进行水面垃圾的提取与识别,并通过SGBM算法进行双目测距,得到距离和角度信息;然后将上位机测得的距离和角度信息通过串口通信发送给下位机Arduino控制板,以控制船体做出转向和航行。测试结果表明,收集装置识别结果稳定,准确率达到90.5%,测距结果准确,精度达到厘米级,能够实现控制装置自动收集的目标。In order to solve the problem of poor location and recognition of garbage target of automatic garbage recovery ship in outdoor waters,a garbage identification and location system of floating garbage recovery ship is proposed,which combines YOLOv8 and binocular ranging algorithm.The system is mainly consisted of three parts:camera,upper computer visual processing unit,and lower computer control unit.After visual identification and classification of garbage in the water environment,location and ranging are carried out to realize the location and identification of garbage in the water area.After positioning and recognition,the machine is controlled to collect garbage.Binocular camera is used to obtain images,Jetson nano embedded chip is used as the host computer main control chip,the latest deep learning model YOLOv8 is used to extract and identify surface garbage,and binocular distance is obtained by means of SGBM algorithm.The distance and angle information measured by the upper computer are sent to the lower computer Arduino control board by the serial communication to control the ship's turning and navigation.The testing results show that the recognition result of the collection device is stable,the accuracy rate can reach 90.5%,the ranging result is accurate,the accuracy can reach centimeter level,and the automatic collection effect of the control device can be achieved.

关 键 词：水面漂浮垃圾 目标定位 垃圾识别 YOLOv8 双目测距算法 视觉检测 自动收集 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391.4[电子电信—信息与通信工程]