基于径赛图像的人体过终点线识别算法研究
|
摘要
随着径赛比赛中运动员竞技水平的不断提高,径赛终点拍摄系统已成为大中型比赛中必不可少的计时设备,该设备的使用为在比赛中得出公平公正的比赛成绩提供了依据和保障。现有径赛终点拍摄系统的比赛成绩是通过终点裁判员在终点冲线图上进行判断,通过手动操作获取的。这就使比赛成绩中引入了人工干预,影响了比赛的公平性；此外,人工成绩判读需要耗费一定的时间,降低了比赛成绩获取的实时性。针对以上问题,本文开展了径赛终点冲线图中人体过终点线识别算法的研究。主要研究内容如下：
首先阐述了现有径赛终点拍摄系统计时工作的基本原理,分析了线阵CCD摄像头的成像机理以及终点冲线图的特点。根据终点冲线图的特点提出了基于行中值的目标检测方法,并与其他几种目标检测方法进行对比。
在目标检测的基础上,利用连通域去除背景波动引入的噪声并实现目标提取。进而对每个提取目标进行头部图像定位,根据所定位的头部图像数目确定所提取的目标中有无人体重叠或粘连情况发生。
对发生重叠或粘连的人体图像,根据头部图像定位结果,提取每个人体图像躯干骨架所在的特征位置。提出了两种依据特征位置进行目标图像分离的方法,欧式距离分离法用于解决重叠较轻或轻度粘连的情况,Floyd分离法用于解决存在交叉重叠的情况。
针对每个独立的人体目标图像轮廓进行分析,与距离变换轮廓分析法相比,采用泊松方程分析法能够得到躯干主体部分的轮廓。在泊松方程分析结果的基础上,提出了一种能够获取轮廓曲线上特征点的特征提取函数,定位特征点后,即可依据特征点位置提取出无损的躯干图像。进而得出冲线位置和比赛成绩。
利用本文提出的基于径赛图像的人体过终点线识别算法,能够在人体目标无重叠或重叠粘连较轻的情况下得出较为准确的比赛成绩。
With the continuous improvement of athletes'competitive level in track race, track end-point timer system has become an essential device in major sporting events and it guarantees exact scores and fair plays.In most existing track end-point timer system, time spent by an athlete is acquired by the operation of umpires on pictures filmed at the finishing line manually. In this way unjust factors may be generated from the inaccurate judgment of umpires, besides, at a time when the game is proceeding rather rapidly, time costed by manual manipulation is indispensable. To solve these problems, this thesis made a research on recogniton algorithm of human body over the finish line. Main contents are as the following:
First, timing principles of existing track end-point timer system is introduced and imaging mechanism of linear CCD used in the system is explained.To realize human object detection,an analyse is made concerning about the features of track images,based on which a method called object detection based on line median is proposed and compared with two other object detection methods.
By connected components abstraction, background noise is removed and human object is extracted. Then the head image of each extracted target is located. According to the number of human head image, determine whether overlap or adhesion occurs in target images.
In the case of overlap or adhesion, trunk skeleton of each human is extracted according to head positions. Two separation methods are proposed. Euclidean distance separation method is for light overlapping or adhesions. Floyd separation method is for crossing overlapping situation.
Silhouette of each single human target is analyzed.Comparing with Distance Transform method, Poisson Equation method can acquire the main contour of human trunk.Based on Poisson Equation method, a feature extraction function is proposed, which can extract non-destructive and accurate trunk image.Based on this, the accurate competition results can be obtained.
In the situation of no or light overlapping and adhesion, recogniton algorithm of human body over the finish line based on track image proposed in this paper can arrive at a accurate result.
首先阐述了现有径赛终点拍摄系统计时工作的基本原理,分析了线阵CCD摄像头的成像机理以及终点冲线图的特点。根据终点冲线图的特点提出了基于行中值的目标检测方法,并与其他几种目标检测方法进行对比。
在目标检测的基础上,利用连通域去除背景波动引入的噪声并实现目标提取。进而对每个提取目标进行头部图像定位,根据所定位的头部图像数目确定所提取的目标中有无人体重叠或粘连情况发生。
对发生重叠或粘连的人体图像,根据头部图像定位结果,提取每个人体图像躯干骨架所在的特征位置。提出了两种依据特征位置进行目标图像分离的方法,欧式距离分离法用于解决重叠较轻或轻度粘连的情况,Floyd分离法用于解决存在交叉重叠的情况。
针对每个独立的人体目标图像轮廓进行分析,与距离变换轮廓分析法相比,采用泊松方程分析法能够得到躯干主体部分的轮廓。在泊松方程分析结果的基础上,提出了一种能够获取轮廓曲线上特征点的特征提取函数,定位特征点后,即可依据特征点位置提取出无损的躯干图像。进而得出冲线位置和比赛成绩。
利用本文提出的基于径赛图像的人体过终点线识别算法,能够在人体目标无重叠或重叠粘连较轻的情况下得出较为准确的比赛成绩。
With the continuous improvement of athletes'competitive level in track race, track end-point timer system has become an essential device in major sporting events and it guarantees exact scores and fair plays.In most existing track end-point timer system, time spent by an athlete is acquired by the operation of umpires on pictures filmed at the finishing line manually. In this way unjust factors may be generated from the inaccurate judgment of umpires, besides, at a time when the game is proceeding rather rapidly, time costed by manual manipulation is indispensable. To solve these problems, this thesis made a research on recogniton algorithm of human body over the finish line. Main contents are as the following:
First, timing principles of existing track end-point timer system is introduced and imaging mechanism of linear CCD used in the system is explained.To realize human object detection,an analyse is made concerning about the features of track images,based on which a method called object detection based on line median is proposed and compared with two other object detection methods.
By connected components abstraction, background noise is removed and human object is extracted. Then the head image of each extracted target is located. According to the number of human head image, determine whether overlap or adhesion occurs in target images.
In the case of overlap or adhesion, trunk skeleton of each human is extracted according to head positions. Two separation methods are proposed. Euclidean distance separation method is for light overlapping or adhesions. Floyd separation method is for crossing overlapping situation.
Silhouette of each single human target is analyzed.Comparing with Distance Transform method, Poisson Equation method can acquire the main contour of human trunk.Based on Poisson Equation method, a feature extraction function is proposed, which can extract non-destructive and accurate trunk image.Based on this, the accurate competition results can be obtained.
In the situation of no or light overlapping and adhesion, recogniton algorithm of human body over the finish line based on track image proposed in this paper can arrive at a accurate result.
引文
[1]Horn B K,Schunk B GDetermining optical flow[J].Artificial Intelligence,1981,17:185-203.
[2]Smith S M,Brady J M.ASSERT-2:Real-time motion segmentation and shape tracking[J].IEEE Trans PAMI,1995,17(7):814-820.
[3]朱克忠.基于光流法对移动目标的视频检测与应用研究[D].合肥：合肥工业大学,2007.
[4]Neri A,Colonnese S,Russo GAutomatic moving object and background separation[J].Signal Processing,1998,66(2):219-232.
[5]Thomas M,Ngan K N.Automatic segmentation of moving objects for video object plane generation[J].IEEE Trans Circuits and Systems for Video Technology,1998,8(5):525-538.
[6]肖自美,肖舟等.约束条件帧间差值运动估值匹配准则[J].中国图像图形学报,1997,2(2,3)：89-93.
[7]C Stauffer,W E L Grimson.Adaptive background mixture models for real-time tracking[C].In:CVPR,IEEE,1999,2:246-252.
[8]Haritaoglu I,Harwood D,Davis L S.W4:real-time surveillance of people and their activities[J].PAMI,2000,22(8):809-830.
[9]C Ridder,O Munkelt,H Kirchner.Adaptive background estimation and foreground detection using Kalman-filtering[C].In:Proc of International Conference on recent Advances in Mechatronics,ICRAM'95,UNESCO Chair on Mechatronics,1995:193-199.
[10]Wren Christopher R,Ali Azarbayejani,Trevor Darrell et al.Pfinder:real-time tracking of the human body[J].PAMI,1997,19(7):780-785.
[11]孙吉花,刘肖琳.一种新的基于统计的背景减除方法[J].计算机工程与应用,2007,43(22)：73-75.
[12]T. Horprasert, D. Harwood, and L.S. Davis. A Statistical Approach for Real-time Robust Background Subtraction and Shadow Detection[J].IEEE International Conference on Computer Vision, Frame-Rate Workshop,1999,1-19.
[13]李庆忠,何东晓.基于聚类的背景建模与运动目标检测方法[J].计算机工程与应用,2008,(8)：73-75.
[14]李仪芳,刘景琳.基于连通域算法的区域测量[J].科学技术与工程,2008,8(9)：2492-2494.
[15]张恒,胡文龙,丁赤飙.基于快速连通域分析的目标特征提取算法[J].计算机工程与应用,2009,45(29)：230-233.
[16]徐正光,鲍东来,张利欣.基于递归的二值图形连通域像素标记算法[J].计算机工程,2006,32(24)：186-189.
[17]高枝宝.基于视频的行人流量检测研究[D].四川大学,2006.
[18]张敏敏.复杂环境下的人头定位技术研究与实现[D].南京理工大学,2008.
[19]邵明明.基于头颈部凹点的人头定位新方法[J].光电子技术,2005,25(2)：100-103.
[20]Hosub Yoon, Dohyung Kim, Suyoung Chi, Youngjo Cho. A robust human head detection method for human tracking[C].Intelligent Robots and Systems,2006 IEEE/RSJ International Conference on 9-15 Oct.2006:4558-4563.
[21]Matsumoto Y,Kato T,Wada T.An Occlusion Robust Likelihood Integration Method for Multi-Camera People Head Tracking[C].Networked Sensing Systems,2007. INSS'07. Fourth International Conference on 6-8 June 2007:235-242.
[22]Yachi K, Wada T, Matsuyama T. Human head tracking using adaptive appearance models with a fixed-viewpoint pan-tilt-zoom camera[C]. Automatic Face and Gesture Recognition, 2000. Proceedings. Fourth IEEE International Conference on.28-30 March 2000 Page(s):150-155.
[23]宋雨潭,纪秀.基于数学形态学二值图像分割算法的研究[J].长春工程学院学报(自然科学版),2008,9(3)：68-70.
[24]吴世雄.卫星图像分割的聚类算法[J].科技信息,2007,2(12)：270-271.
[25]朱玲利,王听忠,鲍苏苏.基于CT图像分析和区域生长法的cr图像分割[J].洛阳理工学院学报(自然科学版),2009,19(4)：55-59.
[26]郭宇春,匡锦瑜.一种纹理图形分割的松弛算法[J].北京师范大学学报(自然科学版),1994,30(1)：60-66.
[27]J K Aggarw,Q Cai.Human motion analysis[J].Computer Vision and Image Understanding. 999,73(3):428-440.
[28]宋福宏.基于汽车人机工程的虚拟人体模型研究[D].哈尔滨工程大学,2007.
[29]吴小毛,马利庄,顾宝军.计算机动画中人体建模与皮肤变形技术的研究现状与展望[J].中国图象图形学报,2007,12(4)：565-573.
[30]Liang Zhao.Dressed human modeling detection and parts localization[D].Carnegie Mellon University,2001.
[31]王娟,庄达民.基于Lagrange方法的失重人体建模与仿真[J].北京航空航天大学学报,2006,32(3)：254-257.
[32]M Oren, C PaPageorgious, P Sinha,E Osuna,T Poggio. Pedestrian detection using wavelet templates[C]. Proceedings of International Conference on Computer Vision and Pattern Recognition,Puerto Rico,1997,193-199.
[33]彭莉,林丽.基于人机系统适配理念的人体模板的设计[J].贵州工业大学学报(自然科学版),2001,30(4)：10-15.
[34]孙庆杰.静态图像中人体检测技术研究[D].中国科学院软件研究所,2003.
[35]Zhilan Hu, Xinggang Lin, Hong Yan. Torso Detection in Static Images[C]. Signal Processing, 2006 8th International Conference on Volume 3,2006:16-20.
[36]Chin-Chang Yu, Hsu-Yung Cheng, Jenq-Neng Hwang, Kuo-Chin Fan. Human body modeling with partial self occlusion from monocular camera[C].Machine Learning for Signal Processing,2008. MLSP 2008. IEEE Workshop on 16-19 Oct.2008:193-198.
[37]L Gorelick, M Galun, E Sharon, R Basri, A Brandt. Shape Representation and Classification Using the Poisson Equation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2006,28(12):1997-2005.
[38]J Shotton, A Blake, R Cipolla.Contour-Based Learning for Object Detection[C].IEEE International Conference Computer Vision,2005:503-510.
[39]张彬,康秀光,陈扶辛.彩色CCD终点计时系统的设计[J].信息技术,2006,14(5)：107-112.
[40]许峰.彩色CCD径赛计时系统研究[D].哈尔滨工程大学,2006.
[41]盛文斌,张威.一种利用邻近原则的聚类方法[J].测绘与空间地理信息,2010,33(1)：202-205.
[42]胡斌,宫宁生.一种改进的Otsu阈值分割算法[J].微电子学与计算机,2009,26(12)：153-155.
[43]刘循,游志胜.多尺度形态学图像边缘检测方法[J].光电工程,2003(3)：56-58.
[44]徐一华,朱玉文.一种人头实时跟踪方法[J].中国图像图形学报,2002,7(1)：11-15.
[45]Gilles Bertrand.A parallel thinning algorithm for medial surface[J].Pattern Recognition Letters,1995,16:979-986.
[46]Tsai Cheng-Fa,Tsai Chun-Wei,Tseng Ching-Chang.A new hybrid heuristic approach for solving large traveling salesman problem[J].Information Sciences,2004,166(1):67-81.
[47]W.P.Choi,K.M.Lam and W.C.Siu.Extraction of the Euclidean Skeleton Based on a Connectivity Criterion[J].Pattern Recognition,2003,36:721-729.
[48]田振夫.求解泊松方程的紧致高阶差分方法[J].西北大学学报,1996(2)：109-114.
[49]田振夫.数值求解泊松方程的四阶紧致差分方法[J].宁夏大学学报,1995(3)：22-24.
[50]马逸尘,梅立泉.偏微分方程现代数值方法[M].北京：科学出版社,2006.
[2]Smith S M,Brady J M.ASSERT-2:Real-time motion segmentation and shape tracking[J].IEEE Trans PAMI,1995,17(7):814-820.
[3]朱克忠.基于光流法对移动目标的视频检测与应用研究[D].合肥：合肥工业大学,2007.
[4]Neri A,Colonnese S,Russo GAutomatic moving object and background separation[J].Signal Processing,1998,66(2):219-232.
[5]Thomas M,Ngan K N.Automatic segmentation of moving objects for video object plane generation[J].IEEE Trans Circuits and Systems for Video Technology,1998,8(5):525-538.
[6]肖自美,肖舟等.约束条件帧间差值运动估值匹配准则[J].中国图像图形学报,1997,2(2,3)：89-93.
[7]C Stauffer,W E L Grimson.Adaptive background mixture models for real-time tracking[C].In:CVPR,IEEE,1999,2:246-252.
[8]Haritaoglu I,Harwood D,Davis L S.W4:real-time surveillance of people and their activities[J].PAMI,2000,22(8):809-830.
[9]C Ridder,O Munkelt,H Kirchner.Adaptive background estimation and foreground detection using Kalman-filtering[C].In:Proc of International Conference on recent Advances in Mechatronics,ICRAM'95,UNESCO Chair on Mechatronics,1995:193-199.
[10]Wren Christopher R,Ali Azarbayejani,Trevor Darrell et al.Pfinder:real-time tracking of the human body[J].PAMI,1997,19(7):780-785.
[11]孙吉花,刘肖琳.一种新的基于统计的背景减除方法[J].计算机工程与应用,2007,43(22)：73-75.
[12]T. Horprasert, D. Harwood, and L.S. Davis. A Statistical Approach for Real-time Robust Background Subtraction and Shadow Detection[J].IEEE International Conference on Computer Vision, Frame-Rate Workshop,1999,1-19.
[13]李庆忠,何东晓.基于聚类的背景建模与运动目标检测方法[J].计算机工程与应用,2008,(8)：73-75.
[14]李仪芳,刘景琳.基于连通域算法的区域测量[J].科学技术与工程,2008,8(9)：2492-2494.
[15]张恒,胡文龙,丁赤飙.基于快速连通域分析的目标特征提取算法[J].计算机工程与应用,2009,45(29)：230-233.
[16]徐正光,鲍东来,张利欣.基于递归的二值图形连通域像素标记算法[J].计算机工程,2006,32(24)：186-189.
[17]高枝宝.基于视频的行人流量检测研究[D].四川大学,2006.
[18]张敏敏.复杂环境下的人头定位技术研究与实现[D].南京理工大学,2008.
[19]邵明明.基于头颈部凹点的人头定位新方法[J].光电子技术,2005,25(2)：100-103.
[20]Hosub Yoon, Dohyung Kim, Suyoung Chi, Youngjo Cho. A robust human head detection method for human tracking[C].Intelligent Robots and Systems,2006 IEEE/RSJ International Conference on 9-15 Oct.2006:4558-4563.
[21]Matsumoto Y,Kato T,Wada T.An Occlusion Robust Likelihood Integration Method for Multi-Camera People Head Tracking[C].Networked Sensing Systems,2007. INSS'07. Fourth International Conference on 6-8 June 2007:235-242.
[22]Yachi K, Wada T, Matsuyama T. Human head tracking using adaptive appearance models with a fixed-viewpoint pan-tilt-zoom camera[C]. Automatic Face and Gesture Recognition, 2000. Proceedings. Fourth IEEE International Conference on.28-30 March 2000 Page(s):150-155.
[23]宋雨潭,纪秀.基于数学形态学二值图像分割算法的研究[J].长春工程学院学报(自然科学版),2008,9(3)：68-70.
[24]吴世雄.卫星图像分割的聚类算法[J].科技信息,2007,2(12)：270-271.
[25]朱玲利,王听忠,鲍苏苏.基于CT图像分析和区域生长法的cr图像分割[J].洛阳理工学院学报(自然科学版),2009,19(4)：55-59.
[26]郭宇春,匡锦瑜.一种纹理图形分割的松弛算法[J].北京师范大学学报(自然科学版),1994,30(1)：60-66.
[27]J K Aggarw,Q Cai.Human motion analysis[J].Computer Vision and Image Understanding. 999,73(3):428-440.
[28]宋福宏.基于汽车人机工程的虚拟人体模型研究[D].哈尔滨工程大学,2007.
[29]吴小毛,马利庄,顾宝军.计算机动画中人体建模与皮肤变形技术的研究现状与展望[J].中国图象图形学报,2007,12(4)：565-573.
[30]Liang Zhao.Dressed human modeling detection and parts localization[D].Carnegie Mellon University,2001.
[31]王娟,庄达民.基于Lagrange方法的失重人体建模与仿真[J].北京航空航天大学学报,2006,32(3)：254-257.
[32]M Oren, C PaPageorgious, P Sinha,E Osuna,T Poggio. Pedestrian detection using wavelet templates[C]. Proceedings of International Conference on Computer Vision and Pattern Recognition,Puerto Rico,1997,193-199.
[33]彭莉,林丽.基于人机系统适配理念的人体模板的设计[J].贵州工业大学学报(自然科学版),2001,30(4)：10-15.
[34]孙庆杰.静态图像中人体检测技术研究[D].中国科学院软件研究所,2003.
[35]Zhilan Hu, Xinggang Lin, Hong Yan. Torso Detection in Static Images[C]. Signal Processing, 2006 8th International Conference on Volume 3,2006:16-20.
[36]Chin-Chang Yu, Hsu-Yung Cheng, Jenq-Neng Hwang, Kuo-Chin Fan. Human body modeling with partial self occlusion from monocular camera[C].Machine Learning for Signal Processing,2008. MLSP 2008. IEEE Workshop on 16-19 Oct.2008:193-198.
[37]L Gorelick, M Galun, E Sharon, R Basri, A Brandt. Shape Representation and Classification Using the Poisson Equation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2006,28(12):1997-2005.
[38]J Shotton, A Blake, R Cipolla.Contour-Based Learning for Object Detection[C].IEEE International Conference Computer Vision,2005:503-510.
[39]张彬,康秀光,陈扶辛.彩色CCD终点计时系统的设计[J].信息技术,2006,14(5)：107-112.
[40]许峰.彩色CCD径赛计时系统研究[D].哈尔滨工程大学,2006.
[41]盛文斌,张威.一种利用邻近原则的聚类方法[J].测绘与空间地理信息,2010,33(1)：202-205.
[42]胡斌,宫宁生.一种改进的Otsu阈值分割算法[J].微电子学与计算机,2009,26(12)：153-155.
[43]刘循,游志胜.多尺度形态学图像边缘检测方法[J].光电工程,2003(3)：56-58.
[44]徐一华,朱玉文.一种人头实时跟踪方法[J].中国图像图形学报,2002,7(1)：11-15.
[45]Gilles Bertrand.A parallel thinning algorithm for medial surface[J].Pattern Recognition Letters,1995,16:979-986.
[46]Tsai Cheng-Fa,Tsai Chun-Wei,Tseng Ching-Chang.A new hybrid heuristic approach for solving large traveling salesman problem[J].Information Sciences,2004,166(1):67-81.
[47]W.P.Choi,K.M.Lam and W.C.Siu.Extraction of the Euclidean Skeleton Based on a Connectivity Criterion[J].Pattern Recognition,2003,36:721-729.
[48]田振夫.求解泊松方程的紧致高阶差分方法[J].西北大学学报,1996(2)：109-114.
[49]田振夫.数值求解泊松方程的四阶紧致差分方法[J].宁夏大学学报,1995(3)：22-24.
[50]马逸尘,梅立泉.偏微分方程现代数值方法[M].北京：科学出版社,2006.