视频对象运动分析与人脸检测试验系统
|
摘要
人脸的检测和识别是众多基于对象的视频应用的技术基础，这些领域包括基于对象的编码和交互、智能人机接口、计算机视觉等。在视频会议、可视电话、身份鉴别、动态场视觉监控等场合中，以人体为主要对象的视频图象越来越多。针对这些领域的应用要求，本论文以视频图象中头肩区域为对象，研究视频对象的运动分析方法、检测静止背景中人体(主要是头肩)的运动方向和速度，实现头部的定位与提取。
本文研究了MPEG中基于块的运动估值与补偿技术，提出一种基于运动矢量统计的运动分析方法。根据最小平均绝对差值(MAD)准则，利用三步搜索算法，计算出相邻两帧视频图象中各子块的运动矢量，该算法通过对运动矢量场中的运动矢量进行统计、分类，找到物体运动的主要方向，而主运动方向上的平均运动矢量就是物体运动的整体矢量。该方法计算简单、运算速度快。
为了确定图象中人脸的位置，本文引入帧间信息处理技术，采用差分分析方法并利用预处理后差分图象的统计特征，提出一种基于二阶微商算子的人脸定位方法。该方法计算人脸在两个投影方向上的统计特征，使用二阶微商算子估计投影曲线斜率的变化，进而找出人脸的外接矩形，判断头部在图象中的位置。该算法不仅运算速度快、定位准确，试验效果好，并且为下一步的工作打下了良好的基础。
本课题的输入数据是使用黑白摄像头和视频图象采集卡得到的数字视频。基于FlyVideo 98-EZ视频采集卡，本文利用VFW(video for windows)技术，编写了视频采集捕获程序，并建立了一个视频数据采集软件平台。该系统可以实现视频信号的纯软件采集、存储、数据格式转换和显示，并为进一步的实际应用提供了功能扩展软件接口。
在上述算法研究的基础上，本文构成了一个基于内容的人脸检测试验系统。针对视频图象人脸检测的应用，该系统可以对视频图象进行实时捕获、存储、预处理，通过运动检测和运动分析计算运动矢量，实现了人脸的自动定位。
Face detection and recognition is the technical foundation of many object-based video applications, such as object-based encoding, decoding and interaction, intelligent human-machine interface, and computer vision. In the videophone, conference call, dynamic monitor and human identification, there are more and more applications of human video object. It is worth studying location and analysis of head in theory and application. Our paper studies the estimation and analyses of partial motion in detail and realize location and extraction of head.
This thesis studies technologies of block-based motion estimation and compensation in MPEQ and gives a method, which is based on motion vector statistics, to calculate motion vector of whole object. According to minimum mean absolute difference criteria (MAD), our paper uses three-step search algorithm to get the block vectors in two sequential images. Main direction of object motion could be obtained by classifying and the average vector on main direction is the vector of whole object.
This thesis starts with an introduction of the motion analysis. Then the processing methods of inter-frame information in video are presented in detail. The paper discusses video processing technical based on difference. After processing of morphologic operations: image eroding and dilating, we get the binary image to every difference image of video. By the reason of characteristic on statistics of images, we gave a face searching and locating algorithm based on the second derivatives operator. We obtain the position of face by calculating the change of projection curve using the second operator. This algorithm reduces computer cost and improves search speed.
The input digital video is from grayscale camera and video capture card. We design video capture software based on drivers of flyvideo EZ capture card. Video can be captured and saved by the software. This program gives interface for other function.
The system can capture and save video. After processing, motion vector can be calculated by means of motion estimate and analysis, and then we can search and locate face.
本文研究了MPEG中基于块的运动估值与补偿技术，提出一种基于运动矢量统计的运动分析方法。根据最小平均绝对差值(MAD)准则，利用三步搜索算法，计算出相邻两帧视频图象中各子块的运动矢量，该算法通过对运动矢量场中的运动矢量进行统计、分类，找到物体运动的主要方向，而主运动方向上的平均运动矢量就是物体运动的整体矢量。该方法计算简单、运算速度快。
为了确定图象中人脸的位置，本文引入帧间信息处理技术，采用差分分析方法并利用预处理后差分图象的统计特征，提出一种基于二阶微商算子的人脸定位方法。该方法计算人脸在两个投影方向上的统计特征，使用二阶微商算子估计投影曲线斜率的变化，进而找出人脸的外接矩形，判断头部在图象中的位置。该算法不仅运算速度快、定位准确，试验效果好，并且为下一步的工作打下了良好的基础。
本课题的输入数据是使用黑白摄像头和视频图象采集卡得到的数字视频。基于FlyVideo 98-EZ视频采集卡，本文利用VFW(video for windows)技术，编写了视频采集捕获程序，并建立了一个视频数据采集软件平台。该系统可以实现视频信号的纯软件采集、存储、数据格式转换和显示，并为进一步的实际应用提供了功能扩展软件接口。
在上述算法研究的基础上，本文构成了一个基于内容的人脸检测试验系统。针对视频图象人脸检测的应用，该系统可以对视频图象进行实时捕获、存储、预处理，通过运动检测和运动分析计算运动矢量，实现了人脸的自动定位。
Face detection and recognition is the technical foundation of many object-based video applications, such as object-based encoding, decoding and interaction, intelligent human-machine interface, and computer vision. In the videophone, conference call, dynamic monitor and human identification, there are more and more applications of human video object. It is worth studying location and analysis of head in theory and application. Our paper studies the estimation and analyses of partial motion in detail and realize location and extraction of head.
This thesis studies technologies of block-based motion estimation and compensation in MPEQ and gives a method, which is based on motion vector statistics, to calculate motion vector of whole object. According to minimum mean absolute difference criteria (MAD), our paper uses three-step search algorithm to get the block vectors in two sequential images. Main direction of object motion could be obtained by classifying and the average vector on main direction is the vector of whole object.
This thesis starts with an introduction of the motion analysis. Then the processing methods of inter-frame information in video are presented in detail. The paper discusses video processing technical based on difference. After processing of morphologic operations: image eroding and dilating, we get the binary image to every difference image of video. By the reason of characteristic on statistics of images, we gave a face searching and locating algorithm based on the second derivatives operator. We obtain the position of face by calculating the change of projection curve using the second operator. This algorithm reduces computer cost and improves search speed.
The input digital video is from grayscale camera and video capture card. We design video capture software based on drivers of flyvideo EZ capture card. Video can be captured and saved by the software. This program gives interface for other function.
The system can capture and save video. After processing, motion vector can be calculated by means of motion estimate and analysis, and then we can search and locate face.
引文
[1] 周洞汝 胡宏斌 等．视频数据库管理系统导论，科学出版社，2000出版
[2] ISO/IEC JTC1/SC29/WG11 N3342, Overview of the MPEG-4 Standard. 2000
[3] 钟玉琢，王琪，、贺玉文。基于对象的多媒体数据压缩编码国际标准MPEG-4及其校验模型。科学出版社，2000出版
[4] 毕厚杰，多媒体信息的传输与处理，人民邮电出版社，1999出版。
[5] ISO/IEC JTC1/SC29/WGll Document N2552 MPEG-4 Video Verification Model 12.1.1998
[6] W Bledsoe. Man-machine facial recognition. Panoramic Research Inc, Palo Alto, CA, 1966,Rep PRI:2
[7] Berto R, Poggio T. Face recognition:Feature versus templates. IEEE Trans. on PAMI, 1993,15(10):1042～1052
[8] Ziquan Hong. Algebraic feature extraction of image for recognition. Pattern Recognition, 1991,24(3):211～219
[9] Nakamura O, Mathur S, Minami T. Identification of human faces based on isodensity maps. Pattern Recognition, 1991,24(3):263～272
[10] L ades M, Vorbuggen J, Buhmann Jet al. Distortion invariant objectrecognition in the dynamic link architecture. IEEE Trans. on Computers, 1991, 42(3):300～31
[11] Samaria F, Young S. HMM-based architecture for face identification. Image and Vision Computing, 1994, 12(8)
[12] L am K M, Yan H. An analytic-to-holistic approach for face recognition based on a single frontal view. IEEE Trans. On PAMI, 1998, 20(7)
[13] L anitis A, Taylor C J, Cootes T F. Automatic interpretation and coding of face images using flexible models. IEEE trans. On PAMI, 1997, 19(7)
[14] A.Murat Tekalp University of Rochester, Digital Video Processing 清华大学出版社，Prentice Hall 公司1997，11
[15] B.K. P. Hork,RobotVision,McGraw-Hall, NewYork,1986
[16] 贾云得，机器视觉，科学出版社，2000
[17] A. MURAT TEKALP著。崔之祜，江春，陈理鑫译，数字视频处理，电子工业出版社，1998
[18] 李智勇，沈振康，杨卫平等，动态图象分析国防出版社 1999
[19] 贾振堂，面向内容视频编码中视频对象分割技术研究
[20] 朱志刚、徐光祜等．自动交通监测的二维时空图象方法．中国图象图形学报，1996。
[21] 艾海舟、乐秀宇．面向视觉监视实时跟踪的动态背景更新方法，计算机工程与应用 2001
[22] 艾海、王栓、何克忠，基于差分图象的人脸检测。中国图象图形学报，1998
[23] Dong Kwon Park,Ho Seok Yoon and Chee Sun Won, Fast Object Tracking In Digital Video, IEEE Transaction on Comsumer Electronics, Vo;46,NO.3,August 2000
[24] 陈朝阳，张桂林，基于图象对称差分运算的运动小目标检测方法，华中理工大学学报 1998
[25] 崔屹，图象处理与分析数字形态学方法及应用，科学出版社，2000
[26] 刘小军、江林华，一种基于投影算法的运动目标自动跟踪系统华北工学院测绘技术学报 1999
[27] 束为，荣钢，边肇祺，张大鹏，基于方向投影的自动掌纹基准点检测，清华大学学报，1998
[28] Yang Guangzheng, Huang T S. Human face detection in complex background. Patt Recog, 1994, 27(1):53～6
[29] Samal A, Iyengar P A. Automatic recognition and analysis of human expression:a survey. Patt Recog, 1992,25(1):1479～1490
[30] Brown M A, Balckwell K T. Multi-scale edge detection and feature binding:an integrated approach. Patt Recog, 1998, 31(10):1479～1490
[31] 施可为，傅锡天，蔡安妮，孙景鳌，含人脸的前景在活动图像序列中的分割 北京邮电大学学报 2000
[32] Ching W S. A novel change detection algorithm using adaptive threshold. Image and Vision Computing, 1994, 12(7):459～463.
[33] Fathy M, Siyal M Y. An image detection technique based on morphological edge detection and background differencing for real-time traffic analysis. Pattern Recognition L ettersl 6,1995:1321～1330.
[34] Jain R. Difference and accumulative difference pictures in dynam-ic scene analysis. Image and Vision Computing, 1984, 2(2):99～10
[35] 王栓 艾海舟 何克忠基于差分图象的多运动目标的检测与跟踪
[36] 艾海舟，王栓，何克忠，基于差分图象的人脸检测 中国图象图形学报
[37] 孙龙祥等，深度图像分析，电子工业出版社，北京，1996。
[38] 黄友谦：曲线曲面的数值表示和逼近，上海科学技术出版社，1984。
[39] 张涛，基于三维表面微分特征的深度图象描述与分割，解放军信息工程大学硕士论文，1999
[40] 赵军，视频信号的捕获采集编程，解放军信息工程大学学报，2001
[2] ISO/IEC JTC1/SC29/WG11 N3342, Overview of the MPEG-4 Standard. 2000
[3] 钟玉琢，王琪，、贺玉文。基于对象的多媒体数据压缩编码国际标准MPEG-4及其校验模型。科学出版社，2000出版
[4] 毕厚杰，多媒体信息的传输与处理，人民邮电出版社，1999出版。
[5] ISO/IEC JTC1/SC29/WGll Document N2552 MPEG-4 Video Verification Model 12.1.1998
[6] W Bledsoe. Man-machine facial recognition. Panoramic Research Inc, Palo Alto, CA, 1966,Rep PRI:2
[7] Berto R, Poggio T. Face recognition:Feature versus templates. IEEE Trans. on PAMI, 1993,15(10):1042～1052
[8] Ziquan Hong. Algebraic feature extraction of image for recognition. Pattern Recognition, 1991,24(3):211～219
[9] Nakamura O, Mathur S, Minami T. Identification of human faces based on isodensity maps. Pattern Recognition, 1991,24(3):263～272
[10] L ades M, Vorbuggen J, Buhmann Jet al. Distortion invariant objectrecognition in the dynamic link architecture. IEEE Trans. on Computers, 1991, 42(3):300～31
[11] Samaria F, Young S. HMM-based architecture for face identification. Image and Vision Computing, 1994, 12(8)
[12] L am K M, Yan H. An analytic-to-holistic approach for face recognition based on a single frontal view. IEEE Trans. On PAMI, 1998, 20(7)
[13] L anitis A, Taylor C J, Cootes T F. Automatic interpretation and coding of face images using flexible models. IEEE trans. On PAMI, 1997, 19(7)
[14] A.Murat Tekalp University of Rochester, Digital Video Processing 清华大学出版社，Prentice Hall 公司1997，11
[15] B.K. P. Hork,RobotVision,McGraw-Hall, NewYork,1986
[16] 贾云得，机器视觉，科学出版社，2000
[17] A. MURAT TEKALP著。崔之祜，江春，陈理鑫译，数字视频处理，电子工业出版社，1998
[18] 李智勇，沈振康，杨卫平等，动态图象分析国防出版社 1999
[19] 贾振堂，面向内容视频编码中视频对象分割技术研究
[20] 朱志刚、徐光祜等．自动交通监测的二维时空图象方法．中国图象图形学报，1996。
[21] 艾海舟、乐秀宇．面向视觉监视实时跟踪的动态背景更新方法，计算机工程与应用 2001
[22] 艾海、王栓、何克忠，基于差分图象的人脸检测。中国图象图形学报，1998
[23] Dong Kwon Park,Ho Seok Yoon and Chee Sun Won, Fast Object Tracking In Digital Video, IEEE Transaction on Comsumer Electronics, Vo;46,NO.3,August 2000
[24] 陈朝阳，张桂林，基于图象对称差分运算的运动小目标检测方法，华中理工大学学报 1998
[25] 崔屹，图象处理与分析数字形态学方法及应用，科学出版社，2000
[26] 刘小军、江林华，一种基于投影算法的运动目标自动跟踪系统华北工学院测绘技术学报 1999
[27] 束为，荣钢，边肇祺，张大鹏，基于方向投影的自动掌纹基准点检测，清华大学学报，1998
[28] Yang Guangzheng, Huang T S. Human face detection in complex background. Patt Recog, 1994, 27(1):53～6
[29] Samal A, Iyengar P A. Automatic recognition and analysis of human expression:a survey. Patt Recog, 1992,25(1):1479～1490
[30] Brown M A, Balckwell K T. Multi-scale edge detection and feature binding:an integrated approach. Patt Recog, 1998, 31(10):1479～1490
[31] 施可为，傅锡天，蔡安妮，孙景鳌，含人脸的前景在活动图像序列中的分割 北京邮电大学学报 2000
[32] Ching W S. A novel change detection algorithm using adaptive threshold. Image and Vision Computing, 1994, 12(7):459～463.
[33] Fathy M, Siyal M Y. An image detection technique based on morphological edge detection and background differencing for real-time traffic analysis. Pattern Recognition L ettersl 6,1995:1321～1330.
[34] Jain R. Difference and accumulative difference pictures in dynam-ic scene analysis. Image and Vision Computing, 1984, 2(2):99～10
[35] 王栓 艾海舟 何克忠基于差分图象的多运动目标的检测与跟踪
[36] 艾海舟，王栓，何克忠，基于差分图象的人脸检测 中国图象图形学报
[37] 孙龙祥等，深度图像分析，电子工业出版社，北京，1996。
[38] 黄友谦：曲线曲面的数值表示和逼近，上海科学技术出版社，1984。
[39] 张涛，基于三维表面微分特征的深度图象描述与分割，解放军信息工程大学硕士论文，1999
[40] 赵军，视频信号的捕获采集编程，解放军信息工程大学学报，2001
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |