摘要
移动机器人作为21世纪最伟大的发明现在越来越多的应用于工业生产和日常生活当中。其中对机器人的视觉研究更是移动机器人领域的核心问题之一。机器视觉的一个主要研究方向就是运动目标的识别与跟踪。在大多数情况下,人们一般只对目标和运动物体感兴趣。比如生产线上对产品的检测,汽车的导航与自动驾驶,无人飞机等等。因此,对运动目标的跟踪的研究有着很大的现实意义和实际价值。
本文以运动的物体为目标,在室内环境下对运动目标进行识别与跟踪。文章主要从图像处理和目标跟踪两方面对移动机器人的目标跟踪进行研究。在图像处理方面,分析现有方法的优缺点,在此基础上提出了一种结合目标形状特征和颜色特征的分割方法。在阈值分割后继续做边缘提取,通过目标的几何特点,计算物体的形状参数,从而识别目标。实验证明此方法可以在复杂的环境下提取我们感兴趣的目标。在目标跟踪过程中,重点研究了Mean-Shift算法并将其用到目标跟踪当中,分析其优缺点并提出将无迹卡尔曼滤波引入Mean-Shift算法当中的方法。通过无迹卡尔曼滤波的预测机制去估计目标下一帧中的位置,在此基础上使用Mean-Shift算法搜索估计区域,得到目标的真实位置,可以大大提高对快速目标跟踪的准确性。实验证明此方法在目标跟踪领域取得了比较理想的效果。
本文对各种设计方法进行了编程实现,所有程序是在XP系统下采用VC++6.0完成的,使用了广茂达ASR机器人自带的函数库和OPENCV函数库,大大简化了编程过程。最后实验表明,上述几种算法运算速度较快,满足系统实时性要求,跟踪目标的准确度较高,满足准确性要求且易于硬件实现,达到预期效果。
Mobile robot, as the greatest invention of the 21st century, is more and more used in industrial production and daily life. Research of robot vision is the core issues of mobile robots. The identification and tracking of moving targets is the main research direction of machine vision. In most cases, people often interested in the target and moving objects. Such as, product testing in the production line, car navigation and automatic driving, unmanned aircraft and so on. Therefore, the study of track for moving target have great practical significance and practical value.
In this paper, the target is movement objects, to identify and tracking moving targets in the indoor. Studying the target tracking of mobile robot mainly from the image processing and target tracking.In image processing, in this based on analyzing strengths and weaknesses of existing methods, it made the segmentation method that combined the features of target shape and color. Continue to do edge detection after the threshold segmentation, through the geometric characteristics of target, calculate shape parameters of the object, and thus identify the target. Experiments proved that this method can extract goal in a complex environment that we are interested in. In the process of target tracking, the focus on the mean-shift algorithm it is used to track target,analyze the algorithm’s strengths and weaknesses and propose the method that introduce the unscented Kalman filter algorithm to the mean-shift. Through the unscented Kalman filter’s prediction mechanism to estimate the target location of the next frame, on this basis, the real target location is kown trough using of mean-shift algorithm to estimate search area, it can greatly enhance the accuracy of the fast-target tracking. Experiments proved that this method can achieve ideal results in the field of target tracking.
In this paper, various design methods were programmed, all programs is achieved in the XP system and used VC + +6.0 , using function library and OPENCV libraries of the Hiroshige ASR robot’s, this greatly simplified the programming process. The last experiments show that the above-mentioned had faster computation speed, met the real-time requirements, had high degree of accuracy to track targets, filled the accuracy requirements and had easy hardware implementation, to achieve the desired results.
引文
[1]李磊,叶涛.移动机器人技术研究现状与未来[J].机器人,2002,24(5):475～480
[2]徐国保,尹怡欣,周美娟.智能移动机器人技术现状及展望[J].机器人技术与应用,2007,2:29～34
[3] M.Sonka, V.Hlavac,R.Boyle. Image Processing Analysis and MachineVision英文版(第三版)[M].北京:人民邮电出版社,2007. 265～276
[4]蔡自兴编著.机器人学[M].北京:清华大学出版社,2000. 113～127
[5]赵海文,岳宏,杜春红.基于主动视觉和超声的机器人目标跟踪系统[J].计算机工程,2008,34(4):220～222
[6]徐国华,谭民.移动机器人的发展现状及其趋势[J].机器人技术与应用,2001,3(5):7～8
[7]李磊,陈细军,候增广.自主轮式移动机器人CASIA-I的整体设计[J].高技术通讯,2007,16(10):265～269
[8]张朋飞,何克忠,欧阳正柱.多功能室外智能移动机器人实验平台-THMR-V[J].机器人,2002, 24(2):97～101
[9]文志强,蔡自兴.基于移动机器人视觉系统的运动目标提取方法[J].计算机工程,2007,24(9):65～69
[10]张俊强,张华.基于图像处理的移动机器人对目标的识别和定位[J].计算机测量与控制,2006,14(5):673～675
[11] Milan Sonka,Vaclav Hlavac,Roger Boyle.图像处理、分析与机器人视觉[M].人民邮电出版社. 2003, 83～90
[12] B.P.Amavasai,F.Caparrelli,A.Selvan,M.Boissenin,J.R.Travis and S.Meikle. Machine vision methods for autonomous micro-robotic systems. Microsystems and Machine Vision Laboratory,Material sand Engineering Research Institute (MERI),School of Engineering, Sheffield Hallam University, Sheffield,UK.2005
[13] Brian Rooks. The vision solution to direct part mark identification. Associate Editor of Sensor Review,2005
[14] Paul G.Ranky.Advanced machine vision systems and application examples.The University of Michigan in Ann Arbor.The Ford Robot Automation and Application Test Center in Michigan, NJIT and Braintech Inc. in the USA.2003.
[15] Christine Connolly. Part-tracking labelling and machine vision. Stalactite Technologies Ltd, HuddersField, UK,2005
[16] T.Pavlidis.36 years on the pattern recognition front:Lecture given at ICPR’2000 in Barcelona,Spain on the occasion of receiving the K.S.Fu prize.Pattern Recognition Letters,2003
[17] J.Shah.Elastica with hinges.Journal of Visual Communication and Image Representation,2002
[18] D.Demigny.On Optimal Linear Filtering for Edge Detection[C]. IEEE Transaction on Image Processing,2002,19(4):321～332
[19]李惠光,任爽,李国友.一种基于颜色信息的足球机器人目标搜索方法[J].计算机仿真,2007,24(3):152～155
[20]韩艳丽,郭晓军,沈同圣.可视化目标特征提取及动态跟踪算法验证系统[J].红外与激光工程,2001,30(2):99～101
[21]侯正信,何宇清,许微.一种快速的图像修复算法[J].中国图像图形学报,2007,10(3):50～53
[22]左飞,万晋森,刘航编著. Visual C++数字图像处理开发入门与编程实践[M].电子工业出版社,2007,103～116
[23]贾永红编著.数字图像处理[M].武汉大学出版社,2003, 78～85
[24]李偲，栾新，宋大雷.基于彩色图像分割的机器人的目标识别[J],微计算机信息,2008,6(2):258～260
[25]张莎莎.图像分割与运动目标跟踪算法研究[D]:[硕士学位论文].哈尔滨:哈尔滨工业大学,2006
[26]杨晓辉,李中科,吴乐南.一种稳健的目标提取与跟踪算法[J].应用科学学报,2005,23(1):31～36
[27]于梅,殷兵,梁栋.基于边缘检测的Contourlet变换图像去噪[J].计算机技术与发展.2008,6:28～31
[28] Milan Sonka,Vaclav Hlavac,Roger Boyle.图像处理分析与机器视觉(第二版)中译本[M].人民邮电出版社,2003,118～136
[29]敬斌.全景视觉足球机器人视觉处理系统设计[D]:[硕士学位论文].西安:西安电子科技大学,2007
[30]卢惠民,王祥科,刘斐.基于全向视觉和前向视觉的足球机器人目标识别[J].中国图像图形学报, 2006,11(11):1686～1689
[31]王建平.移动机器人检测与跟踪系统的研究与设计[D]:[硕士学位论文].合肥:合肥工业大学,2000
[32]朱胜利.Mean Shift及相关算法在视频跟踪中的研究[D]:[博士学位论文].杭州:浙江大学,2006
[33]罗诗途,张玘,罗飞路.基于自适应模板修正的相关跟踪方法[J].仪器仪表学报,2004,1(5):27～31
[34]孙凯,刘士荣.多目标跟踪的改进camshift/卡尔曼滤波组合算法[J].信息与控制,2009,38(1):9～14
[35]汤思维,曹其新.移动机器人多目标彩色视觉跟踪系统[J].机器人,2003,25(1):10～14
[36]杜欣.基于彩色图像的机器人视觉跟踪[D]:[硕士学位论文],北京:北京交通大学,2006
[37]刘祚时,林桂娟,张海英.足球机器人的视觉系统的研制[J].传感器技术,2004,23(3):38～42
[38]杜欣,赵晓光.五自由度立体视觉平台——目标物体的识别与定位[J].计算机工程与应用,2005,41(27):93～96
[39]严国萍,何俊峰.高斯拉普拉斯边缘检测算子的扩展研究[J].华中科技大学武汉学报(自然版),2006,34(10):21～23
[40]陈朝阳,张桂林.基于图像对称差分运算的运动小目标检测方法[J].华中理工大学校报,1998,26(9):34～35
[41]梅跃松.复杂背景下运动目标跟踪关键技术研究[D]:[博士学位论文].北京:北京理工大学,2007
[42]彭亮.机动目标跟踪算法的研究[D]:[硕士学位论文].西安:西北工业大学,2007
[43]童宇,蔡自兴.基于特征匹配的全景图的生成[J].华中科技大学学报(自然科学版),2004,S1:19～22
[44] Dorin Comaniciu.Visvanathan Ramesh,Peter Meer.Real-time tracking of non-grid obiects using mean shift[C]// IEEE International Proceedings of the Computer Vision and Pattern Recognition, 2000,Vo1.2:142～149.
[45]许海霞,王耀南,袁小芳.一种分层Mean Shift目标跟踪算法[J].自动化学报,2009,4(1):9～12
[46] Mean Shift: a robust approach toward feature space analysis ,2002
[47]肖敬若,胡伏原,郑江滨.一种有效的多目标跟踪算法[J].信号处理,2005,z1:348～351
[48]王晋,尚伟,李一民.基于图像处理技术的目标跟踪研究[J].无线电工程,2008,38(7):23～25
[49]陈彬,王浩. CamShift算法在足球机器人系统中的应用与改进[J].合肥工业大学学报(自然科学版),2008,31(7):1074～1077
[50]侯志强,韩崇昭.视觉跟踪技术综述[J].自动化学报,2006,32(4):603～617
[51]栗强,崔慧娟,唐昆.维纳滤波插值算法在运动预测中的应用[J].清华大学学报(自然科学版),2002,42(7):56～59
[52]樊宗臣,吴晓燕.机动目标跟踪的自适应卡尔曼滤波算法实现[J].战术导弹技术,2009,1:79～81
[53]虞旦,韦巍,张远辉.一种基于卡尔曼预测的动态目标跟踪算法研究[J].光电工程,2009,36(1):52～56
[54] Greg Welch,Gary Bishop. The introduction of Kalman Filter. Department of Computer Science University of North Carolina at Chapel Hill,2007
[55]王波,王荫槐,王峰.基于非线性滤波的自适应交互式多模型算法[J].现代雷达,2009,4:51～54
[56]潘泉,杨峰,叶亮.一类非线性滤波器——UKF综述[J].控制与决策,2005,5(6):66～69
[57]朱安福,景占荣,羊彦. UKF及其在目标跟踪中的应用[J].火力与指挥控制,2008,33(8):96～101
[58]贾小军,喻擎苍.基于开源计算机视觉库OpenCV的图像处理[J].计算机应用与软件,2008,25(4):276～278
[59]张苗辉,杨一平,刘先省.基于UKF的机动目标跟踪算法[J].火力与指挥控制,2007,12(8):156～160
[60] Nicolas Saunier and Tarek Sayed. A feature-based tracking algorithmfor vehicles in intersections[C]. IEEE Proceedings of the 3rd Canadian Conference on Computer and Robot Vision,2006,16(3):237～243
[61] Dorin Comaniciu,Visvanathan Ramesh, Peter Meer. Kernel-based object tracking[C].IEEE Trans—actions on Pattern Analysis and Machine Intelligence, 2003, 25(5):564～577
[62] Dorin Comaniciu,Peter Meer.Mean shift:A robust approach toward feature space analysis[C]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2002, 24(5):603～618