基于YOLOv8的足球机器人定位跟踪系统  

A YOLOv8-based soccer robot positioning and tracking system

作　　者：胡欢 邓晗 杨扬 张洛裴 吴运东 梁钰敏 HU Huan;DENG Han;YANG Yang;ZHANG Luopei;WU Yundong;LIANG Yumin(College of Mechanical and Electrical Engineering,Beijing Information Science&Technology University,Beijing 100192,China;College of Automation,Beijing Information Science&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 [2]北京信息科技大学自动化学院,北京100192

出　　处：《北京信息科技大学学报（自然科学版）》2024年第6期88-95,共8页Journal of Beijing Information Science and Technology University（Science and Technology Edition）

基　　金：北京信息科技大学促进高校分类发展-大学生创新创业训练计划项目(5112310851)。

摘　　要：针对RoboCup中型组比赛向人机对抗模式的发展趋势,提出了一种基于YOLOv8的识别多样式足球的方案,并结合双目测距原理进行空间定位,将位姿差异转化为机器人的运动速度和转向角度,实现机器人跟随足球移动。系统整体软件基于机器人操作系统(robot operating system,ROS)搭建,运行于Jetson Xavier NX嵌入式平台上。实验结果表明,基于视觉的足球检测识别算法速度达到56帧/s,能够满足比赛的实时性需求;在对复杂环境验证集的验证中,识别平均精度均值达到了0.767,能够满足比赛的精度要求。With the trend of RoboCup middle size league(MSL)competition evolving towards human-robot confrontation,a versatile soccer recognition scheme based on YOLOv8 was proposed.Combining the principles of binocular ranging for spatial localization,the position difference was transformed into the movement speed and turning angle of the robot,enabling the robot to follow the soccer.The overall software of the system was built on the robot operating system(ROS)and ran on the Jetson Xavier NX embedded platform.Experimental results demonstrate that the vision-based soccer detection and recognition algorithm achieves a speed of 56 frame/s,meeting the real-time requirements of the competition.Moreover,when validated in complex environmental validation sets,the system achieves a mean average precision of 0.767,satisfying the accuracy requirements of the competition.

关 键 词：RoboCup中型组 YOLOv8 目标识别 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]