机器人立体视觉相关技术研究
|
摘要
双目立体视觉是计算机视觉的一个重要分支,它直接模拟人类双眼处理景物的方式,其成果广泛应用于自主式视觉导航、运动目标监视跟踪、工业加工的三维信息获取等领域。本文针对机器人立体视觉相关技术,进行了以下几方面的研究工作：
1、特征提取：特征选用点状特征。点状特征提取采用计算简单,稳定性高的Harris角点检测算法。为了提高角点检测的精度,提出了基于分形插值的二次Harris角点检测方法。这种方法给出了亚像素角点检测的新思路,符合人类观察事物的思维方式。
2、立体匹配：分析了目前常用的匹配方法,总结了各自的优缺点,提出了基于角特征与灰度相关并参考视差梯度约束的角点匹配方法,提高了匹配的速度,实现准确快速的立体匹配。
3、摄像机标定：研究了线性摄像机模型、传统标定方法。针对基于基本矩阵的自标定技术,研究了基本矩阵及其线性求解方法；采用精度高、抗噪声能力强的归一化八点算法求解基本矩阵。
4、运动目标检测与跟踪：静态背景下,提出了改进的PIC背景重构算法,降低了算法的复杂性,提高了程序运行速度。再利用背景差的思想提取出运动目标。动态背景下,研究了基于遗传算法的目标检测方法及不足。针对较为简单的实验目标,采用基于七个不变矩的目标识别方法。最后利用Kalman滤波器提高目标跟踪速度。
5、三维重建：检测出立体图像对中的运动目标,对其质心进行三维重建；控制MOTOMAN-SV3XL工业机器人实现三维空间定位。
Binocular stereovision is an important branch of computer vision, and it directly imitates human manner of binocular to deal with scenery. It is widely used in these filds such as: independence vision navigation, moving object surveillance and tracking, industry artifactitious 3D information acquiring. This paper is to research the robot stereovision technology, and the main work is in the follows aspect:
1、Feature extraction:The point feature is selected as the extracting feature. Harris algorithm for corner detection is used to detect the point feature owing to their fast operation velocity and high precision. In order to get the feature points positioning with a high precision, the dual-Harris corner detection algorithm based on fractal interpolation is proposed.This method accords with human thought manner of observing things.
2、Stereo matching:The matching methods in common use and their advantage and disadvantage are analysed. In order to improve stereo matching time, corner matching method based on the corner characteristic and gray correlation is proposed.
3、Camera calibration:Cameras linear model.traditional calibration method and self-calibrating method based fundational matrix are introduced. The linear algorithm to solve the matrix is researched. Normalized eight-point algorithm is used to estimate the high-precision fundamental matrix.
4、Moving object detection and track:For static background,improved PIC algorithm is proposed to improve the reconstruction speed. And the moving object is detected based on background subtraction. For dynamic background, object detection based on genetic algorithm and its disadvantage is introduced. Then a method for object recognition based on seven invariant moments is researched. At last, Kalman filter is used to improve the real time of object track.
5、3D reconstruction:The 3D reconstruction theory of the binocular stereovision is researched to reconstruct the detected object centroid; MOTOMAN industrial robot is controlled to 3D space position.
1、特征提取：特征选用点状特征。点状特征提取采用计算简单,稳定性高的Harris角点检测算法。为了提高角点检测的精度,提出了基于分形插值的二次Harris角点检测方法。这种方法给出了亚像素角点检测的新思路,符合人类观察事物的思维方式。
2、立体匹配：分析了目前常用的匹配方法,总结了各自的优缺点,提出了基于角特征与灰度相关并参考视差梯度约束的角点匹配方法,提高了匹配的速度,实现准确快速的立体匹配。
3、摄像机标定：研究了线性摄像机模型、传统标定方法。针对基于基本矩阵的自标定技术,研究了基本矩阵及其线性求解方法；采用精度高、抗噪声能力强的归一化八点算法求解基本矩阵。
4、运动目标检测与跟踪：静态背景下,提出了改进的PIC背景重构算法,降低了算法的复杂性,提高了程序运行速度。再利用背景差的思想提取出运动目标。动态背景下,研究了基于遗传算法的目标检测方法及不足。针对较为简单的实验目标,采用基于七个不变矩的目标识别方法。最后利用Kalman滤波器提高目标跟踪速度。
5、三维重建：检测出立体图像对中的运动目标,对其质心进行三维重建；控制MOTOMAN-SV3XL工业机器人实现三维空间定位。
Binocular stereovision is an important branch of computer vision, and it directly imitates human manner of binocular to deal with scenery. It is widely used in these filds such as: independence vision navigation, moving object surveillance and tracking, industry artifactitious 3D information acquiring. This paper is to research the robot stereovision technology, and the main work is in the follows aspect:
1、Feature extraction:The point feature is selected as the extracting feature. Harris algorithm for corner detection is used to detect the point feature owing to their fast operation velocity and high precision. In order to get the feature points positioning with a high precision, the dual-Harris corner detection algorithm based on fractal interpolation is proposed.This method accords with human thought manner of observing things.
2、Stereo matching:The matching methods in common use and their advantage and disadvantage are analysed. In order to improve stereo matching time, corner matching method based on the corner characteristic and gray correlation is proposed.
3、Camera calibration:Cameras linear model.traditional calibration method and self-calibrating method based fundational matrix are introduced. The linear algorithm to solve the matrix is researched. Normalized eight-point algorithm is used to estimate the high-precision fundamental matrix.
4、Moving object detection and track:For static background,improved PIC algorithm is proposed to improve the reconstruction speed. And the moving object is detected based on background subtraction. For dynamic background, object detection based on genetic algorithm and its disadvantage is introduced. Then a method for object recognition based on seven invariant moments is researched. At last, Kalman filter is used to improve the real time of object track.
5、3D reconstruction:The 3D reconstruction theory of the binocular stereovision is researched to reconstruct the detected object centroid; MOTOMAN industrial robot is controlled to 3D space position.
引文
[1]王湘君,周春艳,邹峥嵘.立体视觉概述[J].电脑知识与技术,2005,(36)：145-147
[2]潘春洪,张彩霞.计算机视觉简述[J].机器视觉,2005,(5)：92-93.
[3]Minoru Asada, Takamaro Tanaka. Visual Tracking Of Unkown Moving Object by Adapative Binocular Visual Servoing. Proceeding of the 1999 IEEE. International Conference on Multisensor Fusion and Intelligent Systems.
[4]Yajun Fang, Ichiro Masaki, and Berthold Horn. Depth-Based Target Segmentation for Intelligent Vehicles:Fusion of Radar and Binocular Stereo. IEEE Transanction on Intelligent Transportation Systems,2002,3(3).
[5]Vallerand Steve, Kanbara Masayuki, Yokoya Naokazu. Binocular Vision-Based Augmented Reality System with an Increased Registration Depth Using Dynamic Correction of Feature Positions. Porceedings. Of the 2003 IEEE, Virtual Reality. March 2003, Vol.22-26:p271-272.
[6]Kei Okada, Masayuki Inaba. Integration of Real-Time Binocular Stereo Vision and Whole Body Information for Dynamic Walking Navigation of Humanoid Robot. IEEE Conference on Multisensor Fusion and Intergration for Intelligent Systems,2003.
[7]Clark F. Olson, Habib Abi-Rached, Ming Ye, Jonathan P. Hendrich. Wide-Baseline Stero Vision for Mars Rovers. Proceedings of the 2003 IEEE/RSJ Intl. Conference on Intelligent Robots and Systems. Octobet2003.
[8]隋婧,金伟其.双目立体视觉技术的实现及其进展[J].计算机应用,2004,(10)：4-6.
[9]章毓晋.图像理解与计算机视觉[M].北京：清华大学出版社,2000：81-102,108-141.
[10]邱茂林,马颂德,李毅.计算机视觉中摄像机定标综述[J].自动化学报,2000,26(1)：43-45.
[11]Ganapathy S. Decomposition of transformation matrices for robot vision. In:Proc. Int. Conference on Robotics and Automation.1984,130-139.
[12]Longuet-Higgins HC. A computer algorithm for reconstructing a scene from two projections. Nature,1981,293(10):133-135.
[13]Maybank S J, Faugeras O D. A theory of self-calibration of a moving camera. Intl. J. Computer Vision,1992,8(2):123-151.
[14]Hartley RI. A Linear method for reconstruction from lines and points. In:Proc. ICCV'95.882-887.
[15]潘华,郭戈.立体视觉研究的进展[J].计算机测量与控制,2004,12(12)：1121-1124.
[16]张艳峰.基于立体视觉的不规则物体的三维重建[D].大连：大连理工大学,2005.
[17]C. Harris and M. Stephens. A combined corner and edge detector. In:Alvey Vision Conference, 1988,147-151.
[18]辛菁.机器人无标定视觉伺服控制系统研究[D].西安：西安理工大学,2007.
[19]游素亚,徐光祐.立体视觉研究的现状与进展[J].中国图像图形学报,1997,2(1)：17-24.
[20]徐小明.图像的特征提取与描述的研究[D].重庆：重庆大学,2007.
[21]陈乐,吕文阁,丁少华.角点检测技术研究进展[J].自动化技术与应用,2005,24(5)：1-8.
[22]康晶.基于立体视觉摄像机标定方法的三维重建技术研究[D].湖南：湖南大学,2006.
[23]汪宏异,夏奇.机器视觉中亚象素定位算法研究与改进[J].机床与液压,2004,(10)：165·167.
[24]刘涛,陈国柱,李元宗.亚像素角点定位提高相机标定精度[J].机械工程与自动化,2007,(4)：28-30.
[25]金浩,姜文华,于新瑞,王石刚.基于不同区域的亚像素的插值方法[J].光学仪器,2003,25(4)：7-11.
[26]Kenneth Falconer,曾文曲.分形几何数学基础及其应用[M].北京：人民邮电出版社,2007.
[27]蔡钦涛.基于图像的三维重建技术研究[D].浙江：浙江大学,2004.
[28]Gorodnichy, D Malik, S Roth G. Affordable 3D Face Tracking using Projective Vision[C]. In: Vision Interface 2002, Calgary,2002.
[29]胡晓鹏.双目立体图像的快速匹配值计算方法[J].微计算机信息,2007,(15)：293-294,312.
[30]Hartley, R. I., In defence of the 8-point algorithm, Proceedings of the International Conference on Computer Vision, IEEE Computer Society Press, Boston,1995,1064-1070.
[31]O. Faugeras, F. Lustman. Motion and structure from motion in a piecewise planar environment [J], International Journal of Pattern Recognition and Artificial Intelligence,2(3),1988,485-508.
[32]吴雄君.基于自标定的机器人视觉定位研究[D].西安：西安理工大学,2007.
[33]陈善本,林涛.智能化焊接机器人技术[M].北京：机械工业出版社,2006.
[34]刘传才.图像理解与计算机视觉[M].厦门：厦门大学出版社,2002.
[35]陈小天.摄像机标定技术研究[D].长沙：国防科学技术大学,2003.
[36]钟慧湘.基本矩阵计算方法的研究[D].吉林：吉林大学,2005.
[37]李宏言,盛利元,闻姜.基于改进遗传算法的基础矩阵估计方法[J].计算机工程与应用,2006,(34)：57-59.
[38]胡明星,袁保宗,唐晓芳.基于遗传算法不同策略下的基础矩阵估计方法[J].计算机研究与发展,2002,39(12)：1604-1612.
[39]陈泽志,吴成柯,刘勇.基础矩阵估计的加权归一化线性算法[J].软件学报,2001,12(3)：420-426.
[40]舒远,谈正,丁礼儒.基于高精度匹配点的对极几何估计[J].工程图学学报,2005,(5)：89-92.
[41]杨国亮,王志良.一种改进的光流算法[J].计算机工程,2006,32(15)：187-188.
[42]胡以静,李政访.基于光流的运动分析理论及应用[J].计算机测量与控制,2007,15(2)：219-220.
[43]杨莉.复杂背景下多运动目标轮廓检测[J].电子与信息学报,2005,27(02)：306-309.
[44]关海英,阮秋琦.序列图像中运动目标的检测[J].铁道学报,1996,18(6)：33-38.
[45]崔宇巍.运动目标检测与跟踪中有关问题的研究[D].西安：西北大学,2005.
[46]侯志强,韩崇昭.基于像素灰度归类的背景重构算法[J].软件学报,2005,16(9)：1568·1576.
[47]王小平,曹立明.遗传算法理论、应用与软件实现[M].西安：西安交通大学出版社,2002.
[48]崔宇巍.运动目标检测与跟踪中有关问题的研究[D].西安：西北大学,2005.
[49]朱虹.数字图像处理基础[M].北京：科学出版社,2005.
[50]M. K. Hu, Visual pattern recognition by moment invariants, IEEE Trans. on Information Theory, 1962.8(2):179-187.
[51]杨全.基于7Hu不变矩特征量的中国手指语字母识别算法[D].西安：西安建筑科技大学,2007.
[52]万宜.基于小波神经网络的车牌自动识别研究[D].南京：东南大学,2004.
[53]殷涛.基于几何矩的手势识别算法[D].上海：上海海事大学,2004.
[54]生物医学工程跨系委员会.神经网络及其应用[M].合肥：中国科学技术大学出版社,1992.
[55]程相君,王春宁.神经网络原理及其应用[M].北京：国防工业出版社,1995.
[56]Kalman R E. A New Approach to Linear filtering and Prediction Theory. ASME Journal of Basic Eng,1960,82D:35-46.
[57]王新宇.基于计算机立体视觉的三维重建[D].长沙：中南大学,2004.
[58]MOTOMAN机器人使用说明书[M].北京：首钢莫托曼机器人有限公司,1998.
[59]MOTOMAN机器人操作手册[M].北京：首钢莫托曼机器人有限公司,1998.
[2]潘春洪,张彩霞.计算机视觉简述[J].机器视觉,2005,(5)：92-93.
[3]Minoru Asada, Takamaro Tanaka. Visual Tracking Of Unkown Moving Object by Adapative Binocular Visual Servoing. Proceeding of the 1999 IEEE. International Conference on Multisensor Fusion and Intelligent Systems.
[4]Yajun Fang, Ichiro Masaki, and Berthold Horn. Depth-Based Target Segmentation for Intelligent Vehicles:Fusion of Radar and Binocular Stereo. IEEE Transanction on Intelligent Transportation Systems,2002,3(3).
[5]Vallerand Steve, Kanbara Masayuki, Yokoya Naokazu. Binocular Vision-Based Augmented Reality System with an Increased Registration Depth Using Dynamic Correction of Feature Positions. Porceedings. Of the 2003 IEEE, Virtual Reality. March 2003, Vol.22-26:p271-272.
[6]Kei Okada, Masayuki Inaba. Integration of Real-Time Binocular Stereo Vision and Whole Body Information for Dynamic Walking Navigation of Humanoid Robot. IEEE Conference on Multisensor Fusion and Intergration for Intelligent Systems,2003.
[7]Clark F. Olson, Habib Abi-Rached, Ming Ye, Jonathan P. Hendrich. Wide-Baseline Stero Vision for Mars Rovers. Proceedings of the 2003 IEEE/RSJ Intl. Conference on Intelligent Robots and Systems. Octobet2003.
[8]隋婧,金伟其.双目立体视觉技术的实现及其进展[J].计算机应用,2004,(10)：4-6.
[9]章毓晋.图像理解与计算机视觉[M].北京：清华大学出版社,2000：81-102,108-141.
[10]邱茂林,马颂德,李毅.计算机视觉中摄像机定标综述[J].自动化学报,2000,26(1)：43-45.
[11]Ganapathy S. Decomposition of transformation matrices for robot vision. In:Proc. Int. Conference on Robotics and Automation.1984,130-139.
[12]Longuet-Higgins HC. A computer algorithm for reconstructing a scene from two projections. Nature,1981,293(10):133-135.
[13]Maybank S J, Faugeras O D. A theory of self-calibration of a moving camera. Intl. J. Computer Vision,1992,8(2):123-151.
[14]Hartley RI. A Linear method for reconstruction from lines and points. In:Proc. ICCV'95.882-887.
[15]潘华,郭戈.立体视觉研究的进展[J].计算机测量与控制,2004,12(12)：1121-1124.
[16]张艳峰.基于立体视觉的不规则物体的三维重建[D].大连：大连理工大学,2005.
[17]C. Harris and M. Stephens. A combined corner and edge detector. In:Alvey Vision Conference, 1988,147-151.
[18]辛菁.机器人无标定视觉伺服控制系统研究[D].西安：西安理工大学,2007.
[19]游素亚,徐光祐.立体视觉研究的现状与进展[J].中国图像图形学报,1997,2(1)：17-24.
[20]徐小明.图像的特征提取与描述的研究[D].重庆：重庆大学,2007.
[21]陈乐,吕文阁,丁少华.角点检测技术研究进展[J].自动化技术与应用,2005,24(5)：1-8.
[22]康晶.基于立体视觉摄像机标定方法的三维重建技术研究[D].湖南：湖南大学,2006.
[23]汪宏异,夏奇.机器视觉中亚象素定位算法研究与改进[J].机床与液压,2004,(10)：165·167.
[24]刘涛,陈国柱,李元宗.亚像素角点定位提高相机标定精度[J].机械工程与自动化,2007,(4)：28-30.
[25]金浩,姜文华,于新瑞,王石刚.基于不同区域的亚像素的插值方法[J].光学仪器,2003,25(4)：7-11.
[26]Kenneth Falconer,曾文曲.分形几何数学基础及其应用[M].北京：人民邮电出版社,2007.
[27]蔡钦涛.基于图像的三维重建技术研究[D].浙江：浙江大学,2004.
[28]Gorodnichy, D Malik, S Roth G. Affordable 3D Face Tracking using Projective Vision[C]. In: Vision Interface 2002, Calgary,2002.
[29]胡晓鹏.双目立体图像的快速匹配值计算方法[J].微计算机信息,2007,(15)：293-294,312.
[30]Hartley, R. I., In defence of the 8-point algorithm, Proceedings of the International Conference on Computer Vision, IEEE Computer Society Press, Boston,1995,1064-1070.
[31]O. Faugeras, F. Lustman. Motion and structure from motion in a piecewise planar environment [J], International Journal of Pattern Recognition and Artificial Intelligence,2(3),1988,485-508.
[32]吴雄君.基于自标定的机器人视觉定位研究[D].西安：西安理工大学,2007.
[33]陈善本,林涛.智能化焊接机器人技术[M].北京：机械工业出版社,2006.
[34]刘传才.图像理解与计算机视觉[M].厦门：厦门大学出版社,2002.
[35]陈小天.摄像机标定技术研究[D].长沙：国防科学技术大学,2003.
[36]钟慧湘.基本矩阵计算方法的研究[D].吉林：吉林大学,2005.
[37]李宏言,盛利元,闻姜.基于改进遗传算法的基础矩阵估计方法[J].计算机工程与应用,2006,(34)：57-59.
[38]胡明星,袁保宗,唐晓芳.基于遗传算法不同策略下的基础矩阵估计方法[J].计算机研究与发展,2002,39(12)：1604-1612.
[39]陈泽志,吴成柯,刘勇.基础矩阵估计的加权归一化线性算法[J].软件学报,2001,12(3)：420-426.
[40]舒远,谈正,丁礼儒.基于高精度匹配点的对极几何估计[J].工程图学学报,2005,(5)：89-92.
[41]杨国亮,王志良.一种改进的光流算法[J].计算机工程,2006,32(15)：187-188.
[42]胡以静,李政访.基于光流的运动分析理论及应用[J].计算机测量与控制,2007,15(2)：219-220.
[43]杨莉.复杂背景下多运动目标轮廓检测[J].电子与信息学报,2005,27(02)：306-309.
[44]关海英,阮秋琦.序列图像中运动目标的检测[J].铁道学报,1996,18(6)：33-38.
[45]崔宇巍.运动目标检测与跟踪中有关问题的研究[D].西安：西北大学,2005.
[46]侯志强,韩崇昭.基于像素灰度归类的背景重构算法[J].软件学报,2005,16(9)：1568·1576.
[47]王小平,曹立明.遗传算法理论、应用与软件实现[M].西安：西安交通大学出版社,2002.
[48]崔宇巍.运动目标检测与跟踪中有关问题的研究[D].西安：西北大学,2005.
[49]朱虹.数字图像处理基础[M].北京：科学出版社,2005.
[50]M. K. Hu, Visual pattern recognition by moment invariants, IEEE Trans. on Information Theory, 1962.8(2):179-187.
[51]杨全.基于7Hu不变矩特征量的中国手指语字母识别算法[D].西安：西安建筑科技大学,2007.
[52]万宜.基于小波神经网络的车牌自动识别研究[D].南京：东南大学,2004.
[53]殷涛.基于几何矩的手势识别算法[D].上海：上海海事大学,2004.
[54]生物医学工程跨系委员会.神经网络及其应用[M].合肥：中国科学技术大学出版社,1992.
[55]程相君,王春宁.神经网络原理及其应用[M].北京：国防工业出版社,1995.
[56]Kalman R E. A New Approach to Linear filtering and Prediction Theory. ASME Journal of Basic Eng,1960,82D:35-46.
[57]王新宇.基于计算机立体视觉的三维重建[D].长沙：中南大学,2004.
[58]MOTOMAN机器人使用说明书[M].北京：首钢莫托曼机器人有限公司,1998.
[59]MOTOMAN机器人操作手册[M].北京：首钢莫托曼机器人有限公司,1998.