摘要
随着机器视觉技术的发展,立体视觉尤其是双目立体视觉被广泛应用于物体识别、虚拟现实、工业检测、机器人导航和航空航天等领域。双目立体视觉技术可在多种条件下灵活的获得景物的立体信息,相对单目视觉而言有着不可比拟的优势,是图像处理与机器视觉领域的前沿研究方向。本文对如何在半结构化环境中利用双目立体视觉进行工件的空间三维定位进行了探索与研究。
摄像机标定是机器视觉领域里从二维图像获取三维信息必不可少的步骤。考虑传统标定方法与自标定方法的优点结合具体的实验环境,在不考虑摄像机畸变的情况下,本文采用基于透射投影的双目摄像机线性标定方法,总结了一种直接从图像坐标映射到世界坐标的方法。该方法在保证标定精度的同时,避免了求解摄像机的内参数和外参数带来的误差,大大降低了摄像机标定实验的操作难度和繁琐程度。
为了获得工件的深度信息,对左右摄像机拍摄的图像进行立体匹配。在尺度空间理论基础上,采用图像处理领域的最新研究成果—SIFT特征算子进行匹配,并在具体应用过程中作了改进。实验证明,该方法能很好地解决工件图像具有旋转、缩放、平移、遮挡、噪声等情况下的立体匹配问题,相比原始算法在处理速度上有很大的提高。
在对工件进行识别定位时,利用SIFT特征抵抗旋转、缩放的特点,将其集合作为模板特征在采集的工件图像中寻找对应的工件,并采用基于目标边缘的工件形心计算方法,结合双目摄像机标定实现工件的三维空间定位。在保证算法精度和鲁棒性的前提下,该方法解决了传统模板匹配方法数据量较大的缺点,能够满足实验环境或车间环境下的工件识别定位。
With the development of machine vision, stereo vision especially binocular stereo vision is applied widely in many fields such as objection recognition, virtual reality, industrial inspection, navigator of robotics, aviation and spaceflight. One important task of binocular stereo vision is to get 3D information of objects under various conditions. Binocular stereo vision has incomparable advantages over monocular vision and it is a preceding research area of machine vision. This paper discusses the main search on location of work-piece by binocular stereo vision in semi-structural environment.
Camera calibration is necessary when 3D information is obtained from 2D image in machine vision. Considering conventional calibration methods and self-calibration method and combining concrete experiment environment, binocular linear camera calibration method based on a transmission projection model without distortion is used and the method which image coordinate mapped to world coordinate directly is summarized. This method avoids the error when the intrinsic and extrinsic parameters of the cameras are calculated and makes the calibration experiment much easier with high precision.
Images which are photographed by two cameras are matched in order to get depth information of work-pieces. latest achievement in image processing field—algorithm of Scale Invariant Feature Transform (SIFT) is used to match images based on scale space theory and this algorithm is improved. It is proved by experiments that this improved algorithm can solve stereo matching problems of work-piece images including five aspects: rotation、scale、translation、occlusion and noise and improve processing speed compared to original algorithm.
Scale Invariant Feature Transform (SIFT) is used to search objects as template and calculation method of shape center based on edge is adopted to locate work-pieces combining binocular linear camera calibration. The algorithm in this paper is much more practical than conventional template matching methods because of its high precision and stability.
引文
[1]孙慧.用双目立体视觉技术重建电器触头立体图像[D].河北工业大学,2002.
[2]郭焱,李保国.虚拟植物的研究进展[J].科学通报,2001,46(4):273-280.
[3]李奇,冯华君,徐之海,韩一石,黄红强.计算机立体视觉技术综述[J].光学技术,1999,(5):71-73.
[4]马颂德,张正友.计算机视觉-计算理论与算法基础[M].北京:科学出版社,1998.01.
[5]游素亚,徐光佑.立体视觉研究的现状与进展[J].中国图像图形学报,1997,2(1):16-24.
[6]王天珍.视觉理论与模型的回顾与展望[J].武汉汽车工业大学学报,1996,18(4):29-30.
[7]邱茂林,马颂德,李毅等.计算机视觉中摄像机定标综述[J].自动化学报,2000,26(1):43-55.
[8]袁野.摄像机标定方法及边缘检测和轮廓跟踪算法研究[D].大连理工大学,2002.
[9]刘传才.图像理解与计算机视觉[M].厦门:厦门大学出版社,2005.
[10]Linda G Shapiro,George C Stockman著,赵清杰,钱芳,蔡利栋译.计算机视觉[M].北京:机械工业出版社,2005.
[11]Yakimovsky Y,Cunningham R.A system for extracting three dimensional measurement from a stereo pair of cameras.Computer Graphics and Image Processing,1978,7:195-210.
[12]Hall E.l,Tio M.B.K,McPherson C.A,Sadiadi F.A.Curved surface measurement and recognition for robot vision.In:Proc.IEEE Workshop on Industrial Application of Machine Vision,1982.
[13]Itoh A.M,Ozawa S.Distance measuring methods using only simple vision constructed for moving robots.In:Proc.ICPR'84.1:192-197.
[14]Luh J.Y,Klaasen J.A.A three dimensional vision by off-shelf system with multi-cameras.IEEE Trans.PAMI,1985,7(1).
[15]Tasi R.Y.An efficient and accurate camera calibration technique for 3D machine vision.In:Proc.CVPR'86,364-374
[16]Weng J,Cohen P,Herniou M.Camera calibration with distortion models and accuracy evaluation.IEEE Trans.PAMI,1992,14(10):965-980.
[17]Lenz R K,Tsai R Y.Techniques for calibration of the scale factor and image center for high accuracy 3-D machine vision metrology[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,1988,10(5):713-720.
[18]Martins H.A,Birk J.R,Kelley R.B.Camera models based on data from two calibration planes.Computer Graphics and Imaging Processing,1981,17:173-180.
[19]Ma S,Wei G.A self-calibration technique for active vision system.In:Proc.ACCV'95,628-631.
[20]Ma S.A self-calibration technique for active vision system.IEEE Trans.Robotics and Automation,1996,12:114-120.
[21]O.Faugeras,Q.T.Luong,S.Maybank.Camera Self-Calibration:Theory and Experiments.In Proceedings of the7 the European Conference on Computer Vision,Lecture Notes in Computer Science,Springer-Verlag,1992,321-334.
[22]S.J.Maybank,O.Faugeras.A Theory of Self- Calibration of a Moving Camera.International Journal of Computer Vision,1992,8(2):123-151.
[23]R.I Hartley.Estimation of relative camera position for uncalibrated camera.Proc.of the ECCV'92.Italy:Santa Margherita Ligure,1992,379-387.
[24]ZHOU Ji,SHI Jiao- ying.A robust algorithm for feature point matching[J].Computers &Graphics,2002,26:429-436.
[25]ATALAY V,YILMAZ M U.A matching algorithm based on linear features[J].Pattern Recognition Letters,1998,19:857-867.
[26]GOSHTASBY A,STOCKMAN G C,PAGE C V.A region- based approach to digital image registration with subpixel accuracy[J].IEEE Trans on Geoscience and Remote Sensing,1986,24(3):390-399.
[27]阮秋琦,阮宇智等译.数字图像处理[M].北京:电子工业出版社,2003.3.
[28]Hu M.K.Visual pattern recognition by moment invariants.IEEE Trans.on Info Theory,1962,8(2):179-187.
[29]李博,杨丹,张小洪.基于Harris多尺度角点检测的图像配准新算法[J].计算机工程与应用,2006,(35):37-40.
[30]Iijima T.Basic theory of pattern normalization(for the case of a typical one dimensional pattern)[J].Bulletin of the Electro-technical Laboratory,1962,26:368-388.
[31]Witkin A P.Scale Space Filtering[A].In Int.Joint Conf.Artificial Intelligence,1983:1019-1021.
[32]Koendermk J.The Structure of Image[J].Biological Cybernetics,1984,50:363-370.
[33]Lindeberg T.Scale-Space for discrete Signals[J].IEEE Trans.PAMI,1990,12(3):234-254.
[34]Luis Alvarez,Frederic Guichard,Pierre -Louis Lions,Jean -Michel Morel.Axioms and fundamental equations of image processing[J].Archive for Rational Mechanics and Analysis,200 3(123):199-257
[35]Guillermo Sapiro,Allen Tannenbaum.Area and length preserving geometric invariant scale-spaces.Computer Vision.1994
[36]Luminita A,Vese Tony F,Chan.A Multiphase Level Set Framework for Image Segmentation Using the Mumford and Shah Model[J].International Journal of Computer Vision.2004,3(50):271-293
[37]孙剑,徐宗本.计算机视觉中的尺度空间方法[A].工程数学学报,2005.
[38]Lowe D G.Distinctive image features from scale-invariant key-points[J].International Journal of Computer Vision.2004,60(2):91-110.
[39]李晓明,郑链,胡占义.基于SIFT特征的遥感影像自动配准[J].遥感学报,2006,10(6):885-892.
[40]李寒,牛纪桢,郭禾.基于特征点的全自动无缝图像拼接方法[J].计算机工程与设计,2007,28(9):2082-2085.
[41]Koenderink.The Structure of Images[J].Biological Cybernetics,1984,10:363-396.
[42]黄海赟.立体视觉与三维重建的方法研究[D].上海交通大学,2000.
[43]Z G Zeng,H Yan.Region matching and optimal matching pair theorem[C].In:Computer Graphics International,2001.
[44]T Kanade,M Okutomi.A stereo matching algorithm with an adaptive window[J].Theory and Experiment.1994,PAMI-16(9):920-931.
[45]Kuglin C D,Hines D C.The phase correlation image alignment method[C],processing of IEEE Int Conf on Cybernetics and Society,NewYork.1975:163-165.
[46]Y.Ke,R.Sukthankar.PCA-SIFT:A More Distinctive Representation for Local Image Descriptors.Proc.Conf.Computer Vision and Pattern Recognition,2004,511-517.
[47]张大鹏.模式识别与图像处理并行计算机系统设计[M].哈尔滨工业大学出版社,1998.
[48]黄琛.基于距离模板匹配的行人检测算法与研究[D].清华大学,2006.
[49]尹宏伟.机器视觉技术在烟条检测系统中的应用[D].南京航空航天大学,2007.
[50]汪颖进.目标跟踪过程中遮挡问题的研究[D].华中科技大学,2004.
[51]刘颖.基于分层模板匹配的运动目标跟踪算法研究[D].吉林大学,2007.
[52]何斌.VsiualC++数字图像处理[M].北京:人民邮电出版社,2003.
[53]徐慧.viuslaC++数字图像实用工程案例精选[M].北京:人民邮电出版社,2004.
[54]朱虹.数字图像处理[M].北京:科学出版社,2005.