基于AdaBoost的人脸检测与跟踪算法研究
|
摘要
人脸检测与跟踪是人脸信息处理的先期任务之一,同时是人脸信息处理的关键。人脸图像处理的领域除了包含我们熟知的人脸识别之外,通常还包括表情识别、智能人机交互、视频监控等,而几乎每一个研究方向都离不开人脸检测与跟踪。本文在分析和研究了近年来国内外关于人脸检测与跟踪的最新研究成果的基础之上,对人脸检测与跟踪算法进行了系统的研究,针对人脸非刚性的特点以及复杂环境下的跟踪可能遇到目标容易丢失的问题,提出了基于AdaBoost的实时鲁棒人脸检测与跟踪算法,并在实验室环境下验证了该方法。本文主要研究内容包括：
1、在研究了大量经典的人脸检测算法的基础上,将基于AdaBoost的人脸检测算法与肤色过滤相融合,从而可以减少AdaBoost算法的启动次数,提高了人脸检测的效率；同时,通过删除多余的Harr矩形特征,增加训练样本数据库,从而提高了人脸检测的准确率,降低了人脸误检率。
2、在研究并实际测试了大量跟踪算法的基础上,重点研究和分析了两种人脸跟踪算法：CAMSHIFT和Particle Filter算法。针对人脸跟踪过程中可能出现的复杂背景和遮挡的情况,提出了在正常情况下,使用基于CAMSHIFT全局跟踪器进行全局跟踪；如果人脸信息低于特定阈值,将转用基于Particle Filter的本地跟踪器进行跟踪。
3、利用相关的软硬件环境实际模拟测试了本文中设计的检测与跟踪算法,并将本文提出的一些算法与前人的算法进行了对比和分析,主要对比的是检测与跟踪时间以及实际的检测与跟踪效果。
Face detection and tracking is not only a preliminary task for face information processing, but also a key step to it. In addition to the well-known face recognition, the fields of the face image processing usually include expression recognition, intelligent human-computer interaction and video surveillance, and almost every field is inseparable from the face detection and tracking. This paper based on the latest research results for face detection and tracking at home and abroad, systematic analyses and study the face detection and tracking algorithms. According to problems, such as the face's variability and the tracker may lose the target easily in complicated environment, algorithm based on improved AdaBoost is proposed for face detection and tracking, which is verified real-time performance and robustness under the experimental condition. The main research results of this paper can be concluded as follows:
1. Research a large number of classic face detection algorithms, we combined the face detection algorithm based on AdaBoost with color filter method. By this way, the start-up times of the AdaBoost algorithm goes down and the efficiency of the face detection would be improved. In addition, we reduced the face detection error rate through removing redundant Harr-like rectangle features and inreasing the training sample.
2. Study and actual test a large number of tracking algorithms, we focus on the following two algorithms:the CAMSHIFT and Particle Filter algorithm. According to the complex background and occlusion in the process of face tracking, we put forward that under normal circumstances, the global tracker based on CAMSHIFT is called for tracking, once the face information below a specific threshold value, the local tracker base on Particle Filter is invoked to face tracking.
3. We simulate and test the face detection and tracking algorithm we designed in this paper under certain hardware and software environment, at the same time, we compared the proposed algorithm with previous algorithm in time complexity and the actual effect on the face detection and tracking.
1、在研究了大量经典的人脸检测算法的基础上,将基于AdaBoost的人脸检测算法与肤色过滤相融合,从而可以减少AdaBoost算法的启动次数,提高了人脸检测的效率；同时,通过删除多余的Harr矩形特征,增加训练样本数据库,从而提高了人脸检测的准确率,降低了人脸误检率。
2、在研究并实际测试了大量跟踪算法的基础上,重点研究和分析了两种人脸跟踪算法：CAMSHIFT和Particle Filter算法。针对人脸跟踪过程中可能出现的复杂背景和遮挡的情况,提出了在正常情况下,使用基于CAMSHIFT全局跟踪器进行全局跟踪；如果人脸信息低于特定阈值,将转用基于Particle Filter的本地跟踪器进行跟踪。
3、利用相关的软硬件环境实际模拟测试了本文中设计的检测与跟踪算法,并将本文提出的一些算法与前人的算法进行了对比和分析,主要对比的是检测与跟踪时间以及实际的检测与跟踪效果。
Face detection and tracking is not only a preliminary task for face information processing, but also a key step to it. In addition to the well-known face recognition, the fields of the face image processing usually include expression recognition, intelligent human-computer interaction and video surveillance, and almost every field is inseparable from the face detection and tracking. This paper based on the latest research results for face detection and tracking at home and abroad, systematic analyses and study the face detection and tracking algorithms. According to problems, such as the face's variability and the tracker may lose the target easily in complicated environment, algorithm based on improved AdaBoost is proposed for face detection and tracking, which is verified real-time performance and robustness under the experimental condition. The main research results of this paper can be concluded as follows:
1. Research a large number of classic face detection algorithms, we combined the face detection algorithm based on AdaBoost with color filter method. By this way, the start-up times of the AdaBoost algorithm goes down and the efficiency of the face detection would be improved. In addition, we reduced the face detection error rate through removing redundant Harr-like rectangle features and inreasing the training sample.
2. Study and actual test a large number of tracking algorithms, we focus on the following two algorithms:the CAMSHIFT and Particle Filter algorithm. According to the complex background and occlusion in the process of face tracking, we put forward that under normal circumstances, the global tracker based on CAMSHIFT is called for tracking, once the face information below a specific threshold value, the local tracker base on Particle Filter is invoked to face tracking.
3. We simulate and test the face detection and tracking algorithm we designed in this paper under certain hardware and software environment, at the same time, we compared the proposed algorithm with previous algorithm in time complexity and the actual effect on the face detection and tracking.
引文
[1]李娟.基于DSP的嵌入式人脸识别系统的设计[硕十论文].重庆：重庆大学.2009.
[2]张超.基于视频的远距离人脸检测与识别[硕十论文].北京：北京交通大学.2007.
[3]王映辉.人脸识别——原理、方法与技术[M].北京：科学出版社.2010.
[4]秦晔.人脸检测方法研究与系统实现[硕十论文].南京：南京理工大学.2009.
[5]Turk M A, Pentland A P. Face recognition using eigenfaces[J]. Computer Vision and Pattern Recognition. Proceedings CVPR'91.,Computer Society Conference on. IEEE,1991:586-591.
[6]Yang M H, Abuja N, Kriegman D. Face detection using mixtures of linear subspaces[J]. Automatic Face and Gesture Recognition. Proceedings. Fourth IEEE International Conference on. IEEE,2000:70-76.
[7]Rowley H A, Baluja S, Kanade T. Neural network-based face detection[J]. Pattern Analysis and Machine Intelligence, IEEE Transactions on,1998,20(1):23-38.
[8]Osuna E, Freund R, Girosit F.Training support vector machines:an application to face detection[J]. Computer Vision and Pattern Recognition. Proceedings.,1997 IEEE Computer Society Conference on. IEEE,1997:130-136.
[9]Nefian A V, Hayes Ⅲ M H. An embedded HMM-based approach for face detection and recognition[J]. Acoustics, Speech, and Signal Processing.Proceedings.,1999 IEEE International Conference on. IEEE,1999,6:3553-3556.
[10]Hsu R L, Abdel-Mottaleb M, Jain A K. Face detection in color images[J]. Pattern Analysis and Machine Intelligence, IEEE Transactions on,2002,24(5):696-706.
[11]Viola P, Jones M J. Robust real-time face detection[J]. International journal of computer vision.2004,57(2):137-154.
[12]Zhang C, Zhang Z. A survey of recent advances in face detection[J]. Microsoft Research, June,2010.
[13]Menser B, Muller F. Face detection in color images using principal components analysis[J]. Image Processing and Its Applications. Seventh International Conference on (Conf. Publ. No. 465). IET,1999,2:620-624.
[14]Sobottka K, Pitas I. Face localization and facial feature extraction based on shape and color information[J], Image Processing. Proceedings, International Conference on. IEEE,1996,3: 483-486.
[15]Kruppa H, Castrillon-Santana M, Schiele B. Fast and robust face finding via local context[J], Joint IEEE International Workshop on Visual Surveillance and Performance Evaluation of Tracking and Surveillance (VS-PETS).2003:157-164.
[16]Lerdsudwichai C, Abdel-Mottaleb M, Ansari A. Tracking multiple people with recovery from partial and total occlusion[J], Pattern Recognition,2005,38(7):1059-1070.
[17]Nummiaro K, Koller-Meier E, Van Gool L. An adaptive color-based particle filter[J]. Image and Vision Computing.2003,21(1):99-110.
[18]Ji Q.3D face pose estimation and tracking from a monocular camera[J]. Image and vision computing.2002,20(7):499-511.
[19]韩日升.基于核的变尺度视频目标跟踪算法研究[博士论文].上海：上海交通大学.2009.
[20]Bradski, G., A. Kaehler学习OpenCV(中文版)[M].北京：清华大学出版社.2009.
[21]陈胜勇,刘盛.基于OpenCV的计算机视觉技术实现[M].北京：科学出版社.2008.
[22]朱晓明.基于人脸表情识别的情感研究[硕士论文].浙江：浙江工业大学.2011.
[23]Long P M, Servedio R A. Random classification noise defeats all convex potential boosters[J]. Proceedings of the 25th international conference on Machine learning. ACM, 2008:608-615.
[24]Mansour Y, McAllester D. Boosting using branching programs[J]. Journal of Computer and System Sciences.2002,64(1):103-112.
[25]Schapire R E. The strength of weak leamability[J]. Machine learning.1990,5(2):197-227.
[26]Freund Y, Schapire R, Abe N. A short introduction to boosting[J]. Journal-Japanese Society For Artificial Intelligence.1999,14(771-780):1612.
[27]Freund Y. Boosting a weak learning algorithm by majority[J]. Information and computation. 1995,121(2):256-285.
[28]蔡道英,李学明.基于人脸识别的网络图像管理系统的设计与实现[J].中国科技论文在线.2011.
[29]Lienhart R, Maydt J.An extended set of haar-like features for rapid object detection[J]. Image Processing.Proceedings.2002 International Conference on. IEEE,2002,1:Ⅰ-900-Ⅰ-903 vol. 1.
[30]康寅,杨文川.图像表示及在检索系统中的应用[J].软件.2011,32(1)：38-41.
[31]曹振华.基于视觉的多功能教育机器人设计与实现[硕士论文].苏州：苏州大学.2009.
[32]Comaniciu D, Meer P. Mean shift:A robust approach toward feature space analysis[J]. Pattern Analysis and Machine Intelligence. IEEE Transactions on,2002,24(5):603-619.
[33]江文斌.基于视觉的运动目标检测与跟踪算法研究[硕士论文].杭州：杭州电子科技大学.2012.
[34]Kluge S, Reif K, Brokate M. Stochastic stability of the extended Kalman filter with intermittent observations[J]. Automatic Control. IEEE Transactions on,2010,55(2): 514-518.
[35]Dini D H, Mandic D P, Julier S J. A widely linear complex unscented Kalman filter[J]. Signal Processing Letters. IEEE,2011,18(11):623-626.
[36]朱伟.基于人体匹配解决多人体跟踪中遮挡问题的方法研究[硕士论文].上海：上海交通大学.2009
[37]江艳霞.视频人脸跟踪识别算法研究[博士论文].上海：上海交通大学.2007.
[38]马少平,朱小燕.人工智能[M].北京：清华大学出版社.2001.
[39]屈志毅,周海波.决策树算法的一种改进算法[J].计算机应用.2008,28(6)：141-143.
[40]Bhattacharyya A. On a measure of divergence between two statistical populations defined by their probability distributions[J]. Bull. Calcutta Math. Soc,1943,35(99-109):4.
[41]沈利明.人脸跟踪算法及其在DAM6416平台上的实现[硕士论文].南京：南京理工大学.2006.
[42]牛丽辉.基于预测的Mean Shift人脸跟踪算法研究[硕士论文].福建：华侨大学.2008.
[43]江琦.人脸检测识别与跟踪技术中关键问题的研究[博士论文].吉林：吉林大学.2010.
[44]杨向波,朱虹,邓颖娜,张瞳.基于粒子滤波的人脸跟踪算法[J].计算机工程与应用,2009.44(23)：p.209-211.
[45]王展青,凡友福,张桂林.跟踪遮挡目标的一种鲁棒算法[J].计算机工程与应用.2007,43(027)：50-53.
[2]张超.基于视频的远距离人脸检测与识别[硕十论文].北京：北京交通大学.2007.
[3]王映辉.人脸识别——原理、方法与技术[M].北京：科学出版社.2010.
[4]秦晔.人脸检测方法研究与系统实现[硕十论文].南京：南京理工大学.2009.
[5]Turk M A, Pentland A P. Face recognition using eigenfaces[J]. Computer Vision and Pattern Recognition. Proceedings CVPR'91.,Computer Society Conference on. IEEE,1991:586-591.
[6]Yang M H, Abuja N, Kriegman D. Face detection using mixtures of linear subspaces[J]. Automatic Face and Gesture Recognition. Proceedings. Fourth IEEE International Conference on. IEEE,2000:70-76.
[7]Rowley H A, Baluja S, Kanade T. Neural network-based face detection[J]. Pattern Analysis and Machine Intelligence, IEEE Transactions on,1998,20(1):23-38.
[8]Osuna E, Freund R, Girosit F.Training support vector machines:an application to face detection[J]. Computer Vision and Pattern Recognition. Proceedings.,1997 IEEE Computer Society Conference on. IEEE,1997:130-136.
[9]Nefian A V, Hayes Ⅲ M H. An embedded HMM-based approach for face detection and recognition[J]. Acoustics, Speech, and Signal Processing.Proceedings.,1999 IEEE International Conference on. IEEE,1999,6:3553-3556.
[10]Hsu R L, Abdel-Mottaleb M, Jain A K. Face detection in color images[J]. Pattern Analysis and Machine Intelligence, IEEE Transactions on,2002,24(5):696-706.
[11]Viola P, Jones M J. Robust real-time face detection[J]. International journal of computer vision.2004,57(2):137-154.
[12]Zhang C, Zhang Z. A survey of recent advances in face detection[J]. Microsoft Research, June,2010.
[13]Menser B, Muller F. Face detection in color images using principal components analysis[J]. Image Processing and Its Applications. Seventh International Conference on (Conf. Publ. No. 465). IET,1999,2:620-624.
[14]Sobottka K, Pitas I. Face localization and facial feature extraction based on shape and color information[J], Image Processing. Proceedings, International Conference on. IEEE,1996,3: 483-486.
[15]Kruppa H, Castrillon-Santana M, Schiele B. Fast and robust face finding via local context[J], Joint IEEE International Workshop on Visual Surveillance and Performance Evaluation of Tracking and Surveillance (VS-PETS).2003:157-164.
[16]Lerdsudwichai C, Abdel-Mottaleb M, Ansari A. Tracking multiple people with recovery from partial and total occlusion[J], Pattern Recognition,2005,38(7):1059-1070.
[17]Nummiaro K, Koller-Meier E, Van Gool L. An adaptive color-based particle filter[J]. Image and Vision Computing.2003,21(1):99-110.
[18]Ji Q.3D face pose estimation and tracking from a monocular camera[J]. Image and vision computing.2002,20(7):499-511.
[19]韩日升.基于核的变尺度视频目标跟踪算法研究[博士论文].上海：上海交通大学.2009.
[20]Bradski, G., A. Kaehler学习OpenCV(中文版)[M].北京：清华大学出版社.2009.
[21]陈胜勇,刘盛.基于OpenCV的计算机视觉技术实现[M].北京：科学出版社.2008.
[22]朱晓明.基于人脸表情识别的情感研究[硕士论文].浙江：浙江工业大学.2011.
[23]Long P M, Servedio R A. Random classification noise defeats all convex potential boosters[J]. Proceedings of the 25th international conference on Machine learning. ACM, 2008:608-615.
[24]Mansour Y, McAllester D. Boosting using branching programs[J]. Journal of Computer and System Sciences.2002,64(1):103-112.
[25]Schapire R E. The strength of weak leamability[J]. Machine learning.1990,5(2):197-227.
[26]Freund Y, Schapire R, Abe N. A short introduction to boosting[J]. Journal-Japanese Society For Artificial Intelligence.1999,14(771-780):1612.
[27]Freund Y. Boosting a weak learning algorithm by majority[J]. Information and computation. 1995,121(2):256-285.
[28]蔡道英,李学明.基于人脸识别的网络图像管理系统的设计与实现[J].中国科技论文在线.2011.
[29]Lienhart R, Maydt J.An extended set of haar-like features for rapid object detection[J]. Image Processing.Proceedings.2002 International Conference on. IEEE,2002,1:Ⅰ-900-Ⅰ-903 vol. 1.
[30]康寅,杨文川.图像表示及在检索系统中的应用[J].软件.2011,32(1)：38-41.
[31]曹振华.基于视觉的多功能教育机器人设计与实现[硕士论文].苏州：苏州大学.2009.
[32]Comaniciu D, Meer P. Mean shift:A robust approach toward feature space analysis[J]. Pattern Analysis and Machine Intelligence. IEEE Transactions on,2002,24(5):603-619.
[33]江文斌.基于视觉的运动目标检测与跟踪算法研究[硕士论文].杭州：杭州电子科技大学.2012.
[34]Kluge S, Reif K, Brokate M. Stochastic stability of the extended Kalman filter with intermittent observations[J]. Automatic Control. IEEE Transactions on,2010,55(2): 514-518.
[35]Dini D H, Mandic D P, Julier S J. A widely linear complex unscented Kalman filter[J]. Signal Processing Letters. IEEE,2011,18(11):623-626.
[36]朱伟.基于人体匹配解决多人体跟踪中遮挡问题的方法研究[硕士论文].上海：上海交通大学.2009
[37]江艳霞.视频人脸跟踪识别算法研究[博士论文].上海：上海交通大学.2007.
[38]马少平,朱小燕.人工智能[M].北京：清华大学出版社.2001.
[39]屈志毅,周海波.决策树算法的一种改进算法[J].计算机应用.2008,28(6)：141-143.
[40]Bhattacharyya A. On a measure of divergence between two statistical populations defined by their probability distributions[J]. Bull. Calcutta Math. Soc,1943,35(99-109):4.
[41]沈利明.人脸跟踪算法及其在DAM6416平台上的实现[硕士论文].南京：南京理工大学.2006.
[42]牛丽辉.基于预测的Mean Shift人脸跟踪算法研究[硕士论文].福建：华侨大学.2008.
[43]江琦.人脸检测识别与跟踪技术中关键问题的研究[博士论文].吉林：吉林大学.2010.
[44]杨向波,朱虹,邓颖娜,张瞳.基于粒子滤波的人脸跟踪算法[J].计算机工程与应用,2009.44(23)：p.209-211.
[45]王展青,凡友福,张桂林.跟踪遮挡目标的一种鲁棒算法[J].计算机工程与应用.2007,43(027)：50-53.