视觉追踪机器人系统构建研究  

The construction of visual tracking robot

作　　者：张子洋[1] 孙作雷[1] 曾连荪[1] 

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《电子技术应用》2016年第10期123-126,130,共5页Application of Electronic Technique

基　　金：国家自然科学基金项目(61105097;51279098;61401270);上海市教育委员会科研创新项目(13YZ081)

摘　　要：以低成本机器人Rovio Wow Wee作为下位机平台,通过上位机编程,实现对单目视觉图像和红外数据的在线解析,辨识指定物体,基于前向回溯一致性误差确定待追踪的点,构建视觉追踪机器人系统;并引入在线学习方法,使用追踪、学习和检测算法框架提升了机器人目标追踪的性能。通过PID对机器人进行伺服控制。三组实验初步验证了所提系统的可行性。实验使用C++在Linux下编程实现,并在Github上开源,对科研及市场应用具有一定参考价值。The low cost robot Rovio Wow Wee is employed as the slave machine. With the program deployed on the master ma-chine, the monocular image and infrared data are parsed online, then the specified object is recognized, and the potential feature point is selected by valuing forward- backward consistency error, all these above construct the visual tracking robot system. In order to improve the target tracking performance, the online learning approach, Tracking- Learning- Detection( TLD), is involved. Robot servo control uses PID method. Furthermore, three experiments demonstrate the feasibility of the proposed system. All the demos are implemented by programming on Linux with C + +, the code, which can be freely use for research, is available on Github. It has a certain reference value for the research and application.

关 键 词：视觉追踪 移动机器人 在线学习 PID 开源 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]