基于视觉的机器人动态目标跟踪研究

发布时间：2018-05-18 11:38

本文选题：机器视觉 + 目标跟踪　；参考：《华侨大学》2017年硕士论文

【摘要】：现阶段,越来越多自动化生产线引入工业机器人技术,其中常规作业中多是实现结构化环境里反复的抓取放置过程,灵活性较差。为拓展工业机器人在各个领域的应用,在机器人系统中引入机器视觉,提高对作业环境的自适应能力。本文针对机器人视觉控制系统在动态目标跟踪领域现状进行分析,总结了当前研究的一些不足,如图像算法移植性较差,控制系统开放性较低,跟踪系统不够智能等。在此基础上设计了基于视觉的三轴机器人动态目标跟踪系统。主要研究内容与成果如下:在三轴直线式机械臂上搭建基于PC的视觉控制系统硬件平台,视觉子系统为双相机结构,一个采用眼固定空间的安装方式,控制全局视野,另一个采用眼固定于机器人末端的安装方式,控制局部视野。运动控制子系统采用开放性较高的“运动控制卡+伺服系统”的结构模式。核心图像算法中运动目标检测算法为背景估计法,运动目标跟踪算法为基于SURF的模板匹配。将二者结合可以快速准确的定位动态目标,该算法在目标颜色信息与形状信息未知的情况仍能进行定位跟踪,具有一定通用性。算法通过Halcon机器视觉算法库实现。全局视觉控制系统采用基于位置的控制方式,在MATLAB工具箱中,利用张正友标定法求解内外参数;局部视觉控制系统采用无标定基于图像的控制方式,利用卡尔曼滤波对图像雅克比矩阵进行在线辨识。通过全局与局部控制切换,解决了全局视场被机械臂遮挡后续跟踪无法继续和目标速度运动过快脱离局部视野而跟踪失败的问题,整个过程可完全脱离人工干涉,实现对动态目标的智能化跟踪。编写人机界面软件系统,对整个机器人视觉控制系统方案进行验证,经过实验测试,对目标的跟踪效果达到预期水平,证明本文软硬件方案以及算法的可行性。总结经验与不足,为今后整个视觉控制系统的深入研究与扩展奠定基础。
[Abstract]:At present, more and more automatic production lines are introduced into industrial robot technology, in which the conventional operation is to realize the repeated grasping and placing process in the structured environment, and the flexibility is poor. In order to expand the application of industrial robot in various fields, machine vision is introduced into the robot system to improve the adaptability to the working environment. In this paper, the current situation of robot vision control system in the field of dynamic target tracking is analyzed, and some shortcomings of current research are summarized, such as poor transplantability of image algorithm, low openness of control system, insufficient intelligence of tracking system, and so on. Based on this, a vision-based dynamic target tracking system for a three-axis robot is designed. The main research contents and results are as follows: the hardware platform of PC-based vision control system is built on the three-axis linear manipulator, the vision subsystem is a dual-camera structure, and a fixed eye space is adopted to control the global vision. The other is to control the local field of vision by fixing the eyes at the end of the robot. The motion control subsystem adopts the open mode of motion control card servo system. In the core image algorithm, the moving target detection algorithm is the background estimation method, and the moving target tracking algorithm is the template matching based on SURF. The combination of the two methods can locate the dynamic target quickly and accurately, and the algorithm can still locate and track the dynamic target when the color information and shape information of the target are unknown, so it has a certain generality. The algorithm is implemented by Halcon machine vision algorithm library. The global vision control system adopts a location-based control method. In the MATLAB toolbox, Zhang Zhengyou calibration method is used to solve the internal and external parameters, while the local vision control system adopts an uncalibrated image-based control method. The image Jacobian matrix is identified by Kalman filter. By switching between global and local control, the problem that the global field of view can not continue to be blocked by the robot arm and that the target velocity is too fast out of the local field of vision and the tracking failure can be solved, the whole process can be completely separated from manual interference. Intelligent tracking of dynamic targets is realized. The man-machine interface software system is written to verify the whole robot vision control system. After the experimental test, the tracking effect of the target reaches the expected level, which proves the feasibility of the hardware and software scheme and the algorithm of this paper. Summing up the experience and deficiency, lay a foundation for the further research and expansion of the whole vision control system in the future.
【学位授予单位】：华侨大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【参考文献】