高速缝纫及影像跟踪系统的研究与实现
|
摘要
工业缝纫机零部件的运动轨迹数据可用于数据化对比分析实际行为与设计行为之间的差异,指导缝纫机产品的研发,但该数据难以通过人工观察和记录的方式获取。图像处理方法为此提供了一种新的辅助方式,通过跟踪缝纫机视频图像中的零部件获得运动轨迹数据。目标跟踪是这一过程的关键因素,对分析问题、得出正确结论起着决定性的作用,因此对适用于工业缝纫机运动图像的目标跟踪方法的研究具有重要的应用价值。
本文以高速照相机所拍摄的工业缝纫机运动图像序列为研究对象,设计并实现了高速缝纫机影像跟踪系统,解决了原有系统中不同运动方式下跟踪稳定性不良的问题,实现了缝纫机零部件稳定跟踪的功能。首先,介绍了相关跟踪算法的基本理论。基于对图像序列特点的分析、用户需求的考虑和四种用于灰度图像的相关跟踪算法的仿真比较,提出使用平均平方差相关跟踪算法来完成对零部件的跟踪,实现了短期稳定跟踪。然后,在介绍粒子滤波理论的基础上,将相关跟踪算法嵌入粒子滤波跟踪框架中,采用粒子传播和观测来改进传统相关算法的匹配搜索过程,实现了缝纫机零部件的长期稳定跟踪。最后,引入重采样平滑机制,并改进平滑方式,能够在繁殖大权值粒子的基础上提高粒子的多样性,加强跟踪的稳定性。并经过大量实验确定粒子总数,保持跟踪精度的同时有效减少了跟踪所用时间。
实验结果表明,按上述方法,零部件跟踪的稳定性大大提高,并获得精度高于原系统的零部件运动轨迹,为工业缝纫机的设计研究提供了可靠的分析数据。
The trajectory data of industrial sewing machines are usually used to give a data-based comparative analysis of the differences between actual moving behavior and designed moving behavior, and provide a guide to the research and development work of the sewing product, but the data are difficult to observe and record by artificial means. Image processing method provides a new way for it:obtain the trajectory data by tracking spare parts in the sewing video images. Target tracking is a key factor in this process, and plays a decisive role in analyzing problems and drawing the right conclusion, so the study of target tracking method applicable to high-speed sewing machine motion images has important application value.
In this paper, with the motion image sequence of industrial sewing machines captured by high-speed camera as the research object, High-speed Sewing Machine Image Tracking System is designed and implemented, the problem of the poor tracking stability under different movement pattern in the original system is solved, and a steady tracking is achieved for sewing machine spare parts. First of all, the basic theory of correlation tracking algorithm is introduced. In view of the analysis of the image sequence' characteristics, the consideration for user requirements and the comparison and simulation of four kinds of gray-scale image correlation tracking algorithms, the average squared difference correlation tracking algorithm is used to complete the tracking of spare parts and achieves a short-term stabilization tracking. Then, based on the introduction of the particle filter theory, embed the correlation tracking algorithm into the tracking framework of particle filter, use particle propagation and observation to replace the matching and searching process of the traditional correlation algorithm, and achieve long-term stability of the sewing machine spare parts tracking. Last, the method that it smoothes particles after random sampling is introduced for re-sampling and the smoothing way are improved. This re-sampling way improves the diversity of particles and enhances the stability of tracking with breeding power values particles. And after a large number of experiments the total number of particles is determined, effectively reducing the tracking time on the basis of maintaining tracking accuracy.
Experimental results show that with the above method, the stability of tracking spare parts is greatly improved. More accurate trajectories of components are obtained than that of the original system and reliable analytical data are provided for the study of the sewing machine designing.
本文以高速照相机所拍摄的工业缝纫机运动图像序列为研究对象,设计并实现了高速缝纫机影像跟踪系统,解决了原有系统中不同运动方式下跟踪稳定性不良的问题,实现了缝纫机零部件稳定跟踪的功能。首先,介绍了相关跟踪算法的基本理论。基于对图像序列特点的分析、用户需求的考虑和四种用于灰度图像的相关跟踪算法的仿真比较,提出使用平均平方差相关跟踪算法来完成对零部件的跟踪,实现了短期稳定跟踪。然后,在介绍粒子滤波理论的基础上,将相关跟踪算法嵌入粒子滤波跟踪框架中,采用粒子传播和观测来改进传统相关算法的匹配搜索过程,实现了缝纫机零部件的长期稳定跟踪。最后,引入重采样平滑机制,并改进平滑方式,能够在繁殖大权值粒子的基础上提高粒子的多样性,加强跟踪的稳定性。并经过大量实验确定粒子总数,保持跟踪精度的同时有效减少了跟踪所用时间。
实验结果表明,按上述方法,零部件跟踪的稳定性大大提高,并获得精度高于原系统的零部件运动轨迹,为工业缝纫机的设计研究提供了可靠的分析数据。
The trajectory data of industrial sewing machines are usually used to give a data-based comparative analysis of the differences between actual moving behavior and designed moving behavior, and provide a guide to the research and development work of the sewing product, but the data are difficult to observe and record by artificial means. Image processing method provides a new way for it:obtain the trajectory data by tracking spare parts in the sewing video images. Target tracking is a key factor in this process, and plays a decisive role in analyzing problems and drawing the right conclusion, so the study of target tracking method applicable to high-speed sewing machine motion images has important application value.
In this paper, with the motion image sequence of industrial sewing machines captured by high-speed camera as the research object, High-speed Sewing Machine Image Tracking System is designed and implemented, the problem of the poor tracking stability under different movement pattern in the original system is solved, and a steady tracking is achieved for sewing machine spare parts. First of all, the basic theory of correlation tracking algorithm is introduced. In view of the analysis of the image sequence' characteristics, the consideration for user requirements and the comparison and simulation of four kinds of gray-scale image correlation tracking algorithms, the average squared difference correlation tracking algorithm is used to complete the tracking of spare parts and achieves a short-term stabilization tracking. Then, based on the introduction of the particle filter theory, embed the correlation tracking algorithm into the tracking framework of particle filter, use particle propagation and observation to replace the matching and searching process of the traditional correlation algorithm, and achieve long-term stability of the sewing machine spare parts tracking. Last, the method that it smoothes particles after random sampling is introduced for re-sampling and the smoothing way are improved. This re-sampling way improves the diversity of particles and enhances the stability of tracking with breeding power values particles. And after a large number of experiments the total number of particles is determined, effectively reducing the tracking time on the basis of maintaining tracking accuracy.
Experimental results show that with the above method, the stability of tracking spare parts is greatly improved. More accurate trajectories of components are obtained than that of the original system and reliable analytical data are provided for the study of the sewing machine designing.
引文
[1]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.2003.210-221.
[2]姚剑敏.粒子滤波跟踪方法研究[D].长春：中国科学院研究生院长春光学精密机械与物理研究所博士论文,2005.
[3]D. Comaniciu, V. Ramesh and P. Meer. Kernel-based Object Tracking[J]. IEEE Transaction on Pattern Analysis and Machine intelligence.2003,25(5).546-577.
[4]P. Perez, C. Hue and J. Vermaak, et al. Color-based Probabilistic Tracking[C]. In: Proceedings of the 7th European Conference on Computer Vision.2002,2350. 661-675.
[5]ST. Birchfield and S. Rangarajan. Spatiograms Versus Histograms for Region-based Tracking[C]. In:Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Sam Diego, California, Unite States. 2005,2.1158-1163.
[6]A. Li, Z. Jing and S. Hu. Particle Filter Based Visual Tracking with Multi-cue Adaptive Fusion[J]. Chinese Optics Letters.2005,3(6).326-329.
[7]C. Yang, R. Duraiswami and L. Davis. Fast Multiple Object Tracking via a Hierarchical Particle Filter[C]. In:Proceedings of the 10th IEEE International Conference on Computer Vision. Beijing, Chian.2005,1.212-219.
[8]N. J. B. McFarlane and C. P. Schofield. Segmentation and Tracking of Piglets in Images [J]. Machine Vision and Applications.1995,8(3).187-193.
[9]谢伟蔚.基于纹理特征的单摄像头跟踪研究[D].长沙：湖南大学,2007.
[10]D. DeCarlo and D. Metaxas. Optical Flow Constraints on Deformable Models with Applications to Face Tracking[J]. International Journal of Computer Vision. 2000,38(2).99-127.
[11]A. Murat Tekalp,崔之枯等译.数字视频处理[M].北京：电子工业出版社.1998.
[12]J. Fan, J. Yu and G. Fujita. Spatio-temporal Segmentation for Compact Video Representation[J]. Signal Process:Image Communication.2001,16.553-566.
[13]J. K. Aggarawal and N. Nandhakumar. On the Computaton of Motion from Sequences of Images[J]. In:Proceedings of the IEEE.1988,76.917-935.
[14]R. Merwe, A. Doucet and N. Freitas, et al. The Unscented Particle Filter[D]. Technical report CUED/F-INFENG/TR380. Cambridge University Engineering Department.2000.
[15]Y. Chen,Y. Rui and T. S. Huang. JPDAF Based HMM for Real-time Contour Tracking[C]. In:Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Kauai, Hawaii.2001,1.543-550.
[16]M. Isard and A. Blake. CONDENSATION-Conditional Density Propagation for Visual Tracking[J]. International Journal of Computer Vision.1998,29(1).5-28.
[17]P. Aschwanden, W. Guggenbiihl. Experimental Results from a Comparative Study on Comparative Study on Correlation-type Registration Algorithms[J]. Workshop on Robust Computer Vision. Bonn, ISPRS.1992.268-289.
[18]C. Yang, R. Duraiswami and L. Davis. Efficient Mean-shift Tracking via a New Similarity Measure[C]. In:Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Diego, CA, United States.2005,1.176-183.
[19]彭宁嵩,杨杰,刘志等.Mean-Shift跟踪算法中核函数窗宽的自动选取[J].软件学报.2005,16(9).1542-1551.
[20]S.M.Kay,罗鹏飞等.统计信号处理基础——估计与检测理论[M].北京：电子工业出版社,2003.
[21]J. L. Zhang and J. S. Liu. A New Sequential Importance Sampling Method and its Application to the Two-Dimensional Hydrophobic-Hydrophilic Model[J]. Journal of Chemical Physics.2002,117(7).3492-3498.
[22]C. R. Rao, P. K. Pathak and V. I. Koltchinskii. Bootstrap by Sequential Resampling [J]. Journal of Statistical Planning and Inference.1997,64(2). 257-281.
[23]N. J. Gordon, D. J. Salmond. and A. F. M. Smith. Novel Approach to Nonlinear/Non-Gaussian Bayesian State Estimation [J]. In:IEEE Proceedings-F on Radar and Signal Processing.1993,140(2).107-111.
[24]Christian Heipke. Overview of Image Matching Techniques [EB/OL]. URL:http://phot.epfl.ch/workshop/wks96/art_3_1.html.1996-04-29.
[25]Y. H. Fok and O. C. Au. Improved Fast Matching Algorithm in Feature Domain [J]. In:Proceedings of SPIE:Visual Communications and Image.1994,2308. 1826-1837.
[26]张鹏,卢广山,王合龙等.基于三步搜索法的特征相关目标跟踪算法[J].电光与控制.2004,11(4).38-40.
[27]侯格贤,唐明,马颂德.图象中目标的快速搜索算法[J].中国图象图形学报.2001,6(8).760-764.
[28]张波.基于粒子滤波的图像跟踪算法研究[D].上海：上海交通大学,2007.
[29]张海.基于蒙特卡罗方法的晶体生长模拟系统研究[D].长沙：中南大学,2007.
[30]H. T. Nguyen, Q. Ji and A. W. M. Smeulders. Spatio-temporal Context for Robust Multitarget Tracking[J]. IEEE Transaction on Pattern Analysis and Machine Intelligence.2007,29(1).52-64.
[31]J. Kang, I. Cohen and G. Medioni. Persistent Objects Tracking across Multiple Non-overlapping Cameras. In:Proceedings of IEEE Workshop on Motion and Video Computing. Breckenridge, Colorado, USA.2005,2.112-119.
[32]张卫明,张炎华,钟山.蒙特卡罗粒子滤波算法应用研究[J].微计算机信息.2007,23(1-1).295-297.
[33]S. Godsill and T. Clapp. Improvement Strategies for Monte Carlo Particle Filters [D]. Signal Processing Group, University of Cambridge.
[34]吴宝成.粒子滤波重采样算法研究及其应用[D].哈尔滨：哈尔滨工业大学硕士论文,2006.
[35]张淼,胡建旺,周云峰等.粒子滤波的重采样方法[J].火力与指挥控制.2009,34(10).18-20.
[36]谌剑,严平,张静远.权值优化组合粒子滤波算法研究[J].计算机工程与应用.2009,45(24).33-35.
[37]何立,孙涵,黄永等.MODIS 1B数据的重采样方法研究[J].遥感应用.2007,(3).37-43.
[38]吴刚,唐振民,耿峰等.引入前帧加权采样的粒子滤波目标跟踪算法[J].计算机科学.2009,36(8).215-216,246.
[39]常文革,宋千,梁甸农.重采样差值技术实现与应用[J].系统工程与电子技术.2000,22(5).87-89.
[40]裴福俊,孙新蕊,刘红云等.均匀重采样粒子滤波器在SINS初始对准中的应用[J].北京工业大学学报.2008,34(6).567-571.
[41]高显忠,赵伟,侯中喜.粒子滤波改进算法研究[J].弹箭与制导学报.2009,29(3).241-244.
[42]M. S. Arulampalam, S. Maskell and N. Gordon, et al. A Tutorial on Particle Filters for Online Nonlinear/Non-Gaussian Bayesian Tracking [J]. IEEE Transacions Signal Processing.2002,50(2).174-188.
[43]Y. Rui and Y. Chen. Better Proposal Distributions:Object Tracking using Unscented Particle Filter[C]. In:Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.2001,2.786-793.
[44]陈建良.自回归模型的参数估计[J].山东工业大学学报.1996,26(3).226-230.
[45]范春宇.视频跟踪技术研究[D].西安：西安电子科技大学硕士论文,2007.
[46]赵志国,王首勇,同伟.基于重采样平滑粒子滤波的检测前跟综[J].空军雷达学院学报.2008,22(11).25-28.
[47]胡建旺,张淼.自适应采样粒子滤波算法[J].军械工程学院学报.2009,21(3).56-58.
[48]赵媛媛,叶亮,郭雷.粒子数和采样周期自适应的粒子滤波器[J].计算机工程与应用.2006,42(12).53-56.
[49]孟勃,韩广良,裴炳南等.基于自适应粒子滤波的目标跟踪算法[J].光子学报.2008,37(1).252-255.
[50]章毓晋.图像处理和分析[M].北京：清华大学出版社,1999.245-248.
[2]姚剑敏.粒子滤波跟踪方法研究[D].长春：中国科学院研究生院长春光学精密机械与物理研究所博士论文,2005.
[3]D. Comaniciu, V. Ramesh and P. Meer. Kernel-based Object Tracking[J]. IEEE Transaction on Pattern Analysis and Machine intelligence.2003,25(5).546-577.
[4]P. Perez, C. Hue and J. Vermaak, et al. Color-based Probabilistic Tracking[C]. In: Proceedings of the 7th European Conference on Computer Vision.2002,2350. 661-675.
[5]ST. Birchfield and S. Rangarajan. Spatiograms Versus Histograms for Region-based Tracking[C]. In:Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Sam Diego, California, Unite States. 2005,2.1158-1163.
[6]A. Li, Z. Jing and S. Hu. Particle Filter Based Visual Tracking with Multi-cue Adaptive Fusion[J]. Chinese Optics Letters.2005,3(6).326-329.
[7]C. Yang, R. Duraiswami and L. Davis. Fast Multiple Object Tracking via a Hierarchical Particle Filter[C]. In:Proceedings of the 10th IEEE International Conference on Computer Vision. Beijing, Chian.2005,1.212-219.
[8]N. J. B. McFarlane and C. P. Schofield. Segmentation and Tracking of Piglets in Images [J]. Machine Vision and Applications.1995,8(3).187-193.
[9]谢伟蔚.基于纹理特征的单摄像头跟踪研究[D].长沙：湖南大学,2007.
[10]D. DeCarlo and D. Metaxas. Optical Flow Constraints on Deformable Models with Applications to Face Tracking[J]. International Journal of Computer Vision. 2000,38(2).99-127.
[11]A. Murat Tekalp,崔之枯等译.数字视频处理[M].北京：电子工业出版社.1998.
[12]J. Fan, J. Yu and G. Fujita. Spatio-temporal Segmentation for Compact Video Representation[J]. Signal Process:Image Communication.2001,16.553-566.
[13]J. K. Aggarawal and N. Nandhakumar. On the Computaton of Motion from Sequences of Images[J]. In:Proceedings of the IEEE.1988,76.917-935.
[14]R. Merwe, A. Doucet and N. Freitas, et al. The Unscented Particle Filter[D]. Technical report CUED/F-INFENG/TR380. Cambridge University Engineering Department.2000.
[15]Y. Chen,Y. Rui and T. S. Huang. JPDAF Based HMM for Real-time Contour Tracking[C]. In:Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Kauai, Hawaii.2001,1.543-550.
[16]M. Isard and A. Blake. CONDENSATION-Conditional Density Propagation for Visual Tracking[J]. International Journal of Computer Vision.1998,29(1).5-28.
[17]P. Aschwanden, W. Guggenbiihl. Experimental Results from a Comparative Study on Comparative Study on Correlation-type Registration Algorithms[J]. Workshop on Robust Computer Vision. Bonn, ISPRS.1992.268-289.
[18]C. Yang, R. Duraiswami and L. Davis. Efficient Mean-shift Tracking via a New Similarity Measure[C]. In:Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Diego, CA, United States.2005,1.176-183.
[19]彭宁嵩,杨杰,刘志等.Mean-Shift跟踪算法中核函数窗宽的自动选取[J].软件学报.2005,16(9).1542-1551.
[20]S.M.Kay,罗鹏飞等.统计信号处理基础——估计与检测理论[M].北京：电子工业出版社,2003.
[21]J. L. Zhang and J. S. Liu. A New Sequential Importance Sampling Method and its Application to the Two-Dimensional Hydrophobic-Hydrophilic Model[J]. Journal of Chemical Physics.2002,117(7).3492-3498.
[22]C. R. Rao, P. K. Pathak and V. I. Koltchinskii. Bootstrap by Sequential Resampling [J]. Journal of Statistical Planning and Inference.1997,64(2). 257-281.
[23]N. J. Gordon, D. J. Salmond. and A. F. M. Smith. Novel Approach to Nonlinear/Non-Gaussian Bayesian State Estimation [J]. In:IEEE Proceedings-F on Radar and Signal Processing.1993,140(2).107-111.
[24]Christian Heipke. Overview of Image Matching Techniques [EB/OL]. URL:http://phot.epfl.ch/workshop/wks96/art_3_1.html.1996-04-29.
[25]Y. H. Fok and O. C. Au. Improved Fast Matching Algorithm in Feature Domain [J]. In:Proceedings of SPIE:Visual Communications and Image.1994,2308. 1826-1837.
[26]张鹏,卢广山,王合龙等.基于三步搜索法的特征相关目标跟踪算法[J].电光与控制.2004,11(4).38-40.
[27]侯格贤,唐明,马颂德.图象中目标的快速搜索算法[J].中国图象图形学报.2001,6(8).760-764.
[28]张波.基于粒子滤波的图像跟踪算法研究[D].上海：上海交通大学,2007.
[29]张海.基于蒙特卡罗方法的晶体生长模拟系统研究[D].长沙：中南大学,2007.
[30]H. T. Nguyen, Q. Ji and A. W. M. Smeulders. Spatio-temporal Context for Robust Multitarget Tracking[J]. IEEE Transaction on Pattern Analysis and Machine Intelligence.2007,29(1).52-64.
[31]J. Kang, I. Cohen and G. Medioni. Persistent Objects Tracking across Multiple Non-overlapping Cameras. In:Proceedings of IEEE Workshop on Motion and Video Computing. Breckenridge, Colorado, USA.2005,2.112-119.
[32]张卫明,张炎华,钟山.蒙特卡罗粒子滤波算法应用研究[J].微计算机信息.2007,23(1-1).295-297.
[33]S. Godsill and T. Clapp. Improvement Strategies for Monte Carlo Particle Filters [D]. Signal Processing Group, University of Cambridge.
[34]吴宝成.粒子滤波重采样算法研究及其应用[D].哈尔滨：哈尔滨工业大学硕士论文,2006.
[35]张淼,胡建旺,周云峰等.粒子滤波的重采样方法[J].火力与指挥控制.2009,34(10).18-20.
[36]谌剑,严平,张静远.权值优化组合粒子滤波算法研究[J].计算机工程与应用.2009,45(24).33-35.
[37]何立,孙涵,黄永等.MODIS 1B数据的重采样方法研究[J].遥感应用.2007,(3).37-43.
[38]吴刚,唐振民,耿峰等.引入前帧加权采样的粒子滤波目标跟踪算法[J].计算机科学.2009,36(8).215-216,246.
[39]常文革,宋千,梁甸农.重采样差值技术实现与应用[J].系统工程与电子技术.2000,22(5).87-89.
[40]裴福俊,孙新蕊,刘红云等.均匀重采样粒子滤波器在SINS初始对准中的应用[J].北京工业大学学报.2008,34(6).567-571.
[41]高显忠,赵伟,侯中喜.粒子滤波改进算法研究[J].弹箭与制导学报.2009,29(3).241-244.
[42]M. S. Arulampalam, S. Maskell and N. Gordon, et al. A Tutorial on Particle Filters for Online Nonlinear/Non-Gaussian Bayesian Tracking [J]. IEEE Transacions Signal Processing.2002,50(2).174-188.
[43]Y. Rui and Y. Chen. Better Proposal Distributions:Object Tracking using Unscented Particle Filter[C]. In:Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.2001,2.786-793.
[44]陈建良.自回归模型的参数估计[J].山东工业大学学报.1996,26(3).226-230.
[45]范春宇.视频跟踪技术研究[D].西安：西安电子科技大学硕士论文,2007.
[46]赵志国,王首勇,同伟.基于重采样平滑粒子滤波的检测前跟综[J].空军雷达学院学报.2008,22(11).25-28.
[47]胡建旺,张淼.自适应采样粒子滤波算法[J].军械工程学院学报.2009,21(3).56-58.
[48]赵媛媛,叶亮,郭雷.粒子数和采样周期自适应的粒子滤波器[J].计算机工程与应用.2006,42(12).53-56.
[49]孟勃,韩广良,裴炳南等.基于自适应粒子滤波的目标跟踪算法[J].光子学报.2008,37(1).252-255.
[50]章毓晋.图像处理和分析[M].北京：清华大学出版社,1999.245-248.