【作者】 刘青；

【导师】 王红星； 胡彬生；

【作者基本信息】 南昌工程学院 ， 电子信息（专业学位）， 2023， 硕士

【摘要】 近年机器人技术发展远超人们预期,应用在各个领域,不仅进入工业甚至进入文体,更能进入家庭。其中移动机器人掌握识别追踪本领在体育领域也有实际价值,因此也被重点研究。本文设计了一套基于Kinect的移动机器人目标识别追踪系统,视觉设备选用深度相机Kinect V2、RPLIDAR A1激光雷达用来SLAM和实时避障时使用,选择Kobuki作为移动机器人,在基于Linux的Ubuntu系统下利用ROS、Open CV、Kinect SDK等软件库以及C++编程语言,设计并实现了一款自动识别跟踪移动机器人。主要完成的任务有:(1)根据实际应用需求来确定移动机器人要具备的应用功能,并以此设计了总体系统硬件结构各个组成和所需的软件模块。(2)针对小目标的检测进行优化。通常提高小目标检测最直接的方法是采用更高的分辨率图像,但也导致计算成本随着图像像素的增加而急剧地增长,并且产生了更高的冗余计算。为此设计了一种基于Query Det-Yolo模型,使用传感器Kinect的实时检测定位系统。Kinect在ROS系统下利用Open CV,通过查询机制和映射关系,首先预测小目标物体在低分辨率特征上的大概位置,然后使用由那些预测位置在HD下稀疏卷积来计算精确检测结果,达到了减少在直接高分辨率下检测计算量过大的问题,加快了检测速度,同时提高了检测效果,配合点云,输出三维坐标信息,达到了实时精确检测定位目标的效果。进行消融实验,证实改进模块的有效性;进行对比实验,用COCO数据集和Visdrone数据集和其他算法进行比较,证实Query Det算法的优越性;最后针对乒乓球,采集数据集,进行针对性训练,以提升实验效果。(3)设计追踪策略,针对目标乒乓球采用目标识别定位系统提供的坐标信息转换成终点目标,然后从全局路径规划和局部避障两个方面设计了导航避障模块。针对全局路径规划算法,设计了一种基于A*算法的全局规划路线,通过ROS功能包,实现了稳定、实时的路径规划功能;针对局部避障算法,提出了一种移动窗口法(DWA),实现了对于移动障碍物的实时避障功能,保证了移动机器人运行的安全性。将各模块整合成一个系统,分别对各模块和整体系统进行实验测试,实验结果表明本文设计的识别追踪机器人的性能达到了预期设计目标。更多还原

【Abstract】 Recent developments in robotics have far exceeded expectations,with applications in a variety of fields,not only in industry and even in sport and culture,but also in the home.In recent years,mobile robots have been used in a variety of applications,not only in industry and even in sport,but also in the home.This paper designs a Kinectbased mobile robot target recognition and tracking system,using Kinect V2 and RPLIDAR A1 radar as propriety sensors,Kobuki as robot chassis,and software libraries such as ROS,Open CV,Kinect SDK and C++ programming language under Linux system,to design and implement an intelligent table tennis automatic ball picking robot.The main work of this paper includes:(1)analysing the requirements and determining the target functions to be achieved by the robot,and using this to design the various components of the overall system hardware architecture and the required software modules.(2)Optimisation of the detection of small object.The most direct way to improve small object detection is usually to use higher resolution images,but this also leads to a dramatic increase in computational cost as the number of image pixels increases,and creates a higher level of redundant computation.To this end,a real-time detection and localisation system based on the Query Det-Yolo model was designed using the sensor Kinect,which uses Open CV under ROS to first predict the approximate position of small object on low-resolution features through query mechanisms and mapping relationships,and then use sparse convolution from those predicted positions under HD to compute accurate detection The results achieve a reduction in the problem of overcomputation of detection at direct high resolution,speeding up detection while improving detection,and with point clouds,outputting 3D coordinate information to achieve accurate detection of localised targets in real time.Ablation experiments are conducted to confirm the effectiveness of the improved module;comparison experiments are conducted to compare the COCO dataset with the Visdrone dataset and other algorithms to confirm the superiority of the Query Det algorithm;finally,for pingpong balls,datasets are collected and targeted training is carried out to enhance the experimental effect.(3)To design the tracking strategy,the coordinate information provided by the target recognition and positioning system is used to convert the target into an end-point target for the target table tennis ball,and then the navigation and obstacle avoidance module is designed in terms of both global path planning and local obstacle avoidance.For the global path planning algorithm,a global planning route based on the A*algorithm is designed to achieve a stable and real-time path planning function through the ROS function package;for the local obstacle avoidance algorithm,a moving window method(DWA)is proposed to achieve a real-time obstacle avoidance function for moving obstacles and ensure the safety of the mobile robot operation.The modules are integrated into one system,and experimental tests are conducted on each module and the overall system respectively.The experimental results show that the performance of the identification and tracking robot designed in this paper meets the expected design objectives.更多还原