基于CPU+GPU桌面集群的人脸特征点实时检测系统研究
|
摘要
人脸特征检测作为模式识别和机器视觉的一个重要研究方向,在身份识别、基于内容的检索、新一代人机交互等应用领域都得到了广泛的研究。由于这些应用都具有实时性的约束,因此如何加快人脸特征检测的速度就成为了人脸检测方面研究的重要课题。计算统一设备架构(CUDA ,Compute Unified Device Architecture)开启了使用GPU强大计算能力做通用计算的大门,使得开发者能够在友好的开发环境中充分挖掘GPU的计算能力,使得基于CPU+GPU的高效计算模式成为可能,为实现人脸特征检测的实时性提供了新的有效技术支持。
本文根据图像处理算法及人脸特征检测算法,提出了一种基于CPU+GPU桌面集群的人脸特征检测系统,利用GPU的高数据并行计算能力及其内部高数据带宽,实现流多处理器间的粗粒度任务级级并行,以及流多处理器内的细粒度数据并行,从而极大地提高算法的运行速度,以实现对人脸特征的实时检测。其中,将处理任务进行合理的划分(将算法中耗费时间多的大规模数据并行、高计算密度的部分映射到GPU端,算法中的其他部分则映射到CPU端)以及GPU端的任务并行性设计是本文的主要研究内容。
本文将该系统在GTX260上进行了验证,对于分辨率为640*480的24位BMP位图,图像处理速度可达71ms/f,检测率为94% ,达到了相关应用的实时性要求,为基于桌面系统的各种实时性人机交互领域提供了有效支持。
As an important research topic of the pattern recognition and machine vision, the human face feature detection technology has been studied widely in the application area such as the face recognition, new human-computer interaction, information security etc. For these applications have the limitation of the real-time, how to accelerate the speed of the face feature detection has always been an important topic. While CUDA (Compute Unified Device Architecture) starts a new area of doing general-purpose computing on GPUs by providing developers a friendly development environment to fully use GPU’s computing power, it also makes the high CPU+GPU cooperative computation model possible. So it provides a new implementation scheme of the real-time face feature detection for us.
In this paper, we developed a CPU+GPU desktop face feature detection system which uses the high data-parallel computing power and the high internal data bandwidth of GPU to achieve the coarse granularity task level parallelism and the fine granularity data level parallelism, Thus greatly improve the speed of the algorithm to realize the real-time face feature points detection. Specially, the main work of this dissertation includes the reasonable task assignment of the algorithm to take full advantage of GPU and effectively balances the workload between CPU and GPU -- part of the high data parallel and enormous computation in the algorithm will be mapped into the platform of GPU and the remainder will be mapped into the Platform of CPU—and the design of the parallelism of the algorithm which will be mapped into the platform of GPU. Finally, we test and verify the system running on a NVIDIA Gefoce GTX260 graphics card, which achieve the speed of 71ms/f for a BMP image which size is 640*480 and the detection rate of 94%. Our system can meet real-time acquirement of the relevant applications and provide effective support for all kinds of real-time human-computer interaction field based on the desktop system.
本文根据图像处理算法及人脸特征检测算法,提出了一种基于CPU+GPU桌面集群的人脸特征检测系统,利用GPU的高数据并行计算能力及其内部高数据带宽,实现流多处理器间的粗粒度任务级级并行,以及流多处理器内的细粒度数据并行,从而极大地提高算法的运行速度,以实现对人脸特征的实时检测。其中,将处理任务进行合理的划分(将算法中耗费时间多的大规模数据并行、高计算密度的部分映射到GPU端,算法中的其他部分则映射到CPU端)以及GPU端的任务并行性设计是本文的主要研究内容。
本文将该系统在GTX260上进行了验证,对于分辨率为640*480的24位BMP位图,图像处理速度可达71ms/f,检测率为94% ,达到了相关应用的实时性要求,为基于桌面系统的各种实时性人机交互领域提供了有效支持。
As an important research topic of the pattern recognition and machine vision, the human face feature detection technology has been studied widely in the application area such as the face recognition, new human-computer interaction, information security etc. For these applications have the limitation of the real-time, how to accelerate the speed of the face feature detection has always been an important topic. While CUDA (Compute Unified Device Architecture) starts a new area of doing general-purpose computing on GPUs by providing developers a friendly development environment to fully use GPU’s computing power, it also makes the high CPU+GPU cooperative computation model possible. So it provides a new implementation scheme of the real-time face feature detection for us.
In this paper, we developed a CPU+GPU desktop face feature detection system which uses the high data-parallel computing power and the high internal data bandwidth of GPU to achieve the coarse granularity task level parallelism and the fine granularity data level parallelism, Thus greatly improve the speed of the algorithm to realize the real-time face feature points detection. Specially, the main work of this dissertation includes the reasonable task assignment of the algorithm to take full advantage of GPU and effectively balances the workload between CPU and GPU -- part of the high data parallel and enormous computation in the algorithm will be mapped into the platform of GPU and the remainder will be mapped into the Platform of CPU—and the design of the parallelism of the algorithm which will be mapped into the platform of GPU. Finally, we test and verify the system running on a NVIDIA Gefoce GTX260 graphics card, which achieve the speed of 71ms/f for a BMP image which size is 640*480 and the detection rate of 94%. Our system can meet real-time acquirement of the relevant applications and provide effective support for all kinds of real-time human-computer interaction field based on the desktop system.
引文
[1]朱葛俊,张力,盛昀瑶多核系统的多线程并行计算分析南京工业职业技术学院学报7.2009
[2]李刚,高政、人脸自动识别方法综述、计算机应用研究、2003, Vol 8:4-9
[3] C.kotropoulos, I. Pitas. Rule-Based Face Detection in Frontal Views. Proc. Of Int '1 Conf. Acoustics, Speech and Signal Processing. 1997:25(4)
[4] G.Yang, T.S. Huang. Human Face Detection in Complex Background. Patten Recognition.1994, 27(l1):53-6
[5] D.Chetverikov, A.Lerch. Multiresolution Face Detection. Theoretical Foundations of Computer Vision. 1993, 69: 131-140
[6] H.P. Graf, T. Chen, E. Petajan, E. Cosatto. Locating Faces and Facial Parts. Proc. First Int’l Workshop Automatic Face and Gesture Recognition. 1995: 41-46
[7] T.K. Leung, M.C. Burl, P. Perona. Finding Faces in Cluttered Scenes Using Random Labeled Graph Matching. Proc. Fifth IEEE Int’l Conf. Computer Vision. 1995: 637-644
[8] K.C. Yow, R. Cipolla. A Probabilistic Framework for Perceptual Grouping of Features for Human Face Detection. Proc. Second Int’l Conf. Automatic Face and Gesture Recognition. 1996: 16-21
[9] Augusteijn MF,Skufca TL.Identification of human faces through texture-based feature recognition and neural network technology[C].Proc IEEE Int Joint Conf Neural Networks 1993:392–398.
[10] Y.Dai,Y.Nakano. Face-Texture Model Based on SGLD and Its Application in Face Detection in a Color Scene[C].Pattern Recognition.1996,V01.29,no.6:1007-1017.
[11] Yoo T W, Oh I S.A fast algorithm for tracking human faces based on chromatic histograms[C].Pattern Recognition Letters,1999,20(10):967-978.
[12] Moghaddam B,Pentland A.Probabilistic visual learning for object representation [C].IEEE Trans Pattern Analysis and Machine Intelligence,1997,19(7):696-710.
[13] Rowley H A,Baluja S,Kanade T.Rotation invariant neural network-based face detection[Z].Carnegie Mellon University,Pittsburgh PA:Technical Report CMU-CS-97-201,1997.
[14] Osuna E,Freund R,Girosi F.Training support vector machines:An application to face detection[C].In:Proc Computer Vision and Pattern Recognition,Puerto Rico,1997:130-136.
[15] Nefian A V,Hayes M H.Face detection and recognition using hidden Markov models[C].In:Proc IEEE Conference on Image Processing,Chicago,1998:141-145.
[16]张舒,褚艳利,赵开勇,GPU高性能运算之CUDA,中国水利水电出版社2009 36-58 120-140
[17] NVIDIA CUDA Best Practices Guide, NVIDIA,Version3.0 Apr. 2010.[Online].Available:http://cuda.csdn.net/download/CUDA Toolkit version 1.1 for Windows XP/C/doc/NVIDIA_CUDA_BestPracticesGuide.pdf
[18] Schapire R. E. A Brief Introduction of Boosting. Proceedings of the 16th International Joint Conference on Artificial Intelligence. 1999
[19] Viola P and Jones M. Rapid Object Detection using a Boosted Cascade of Simple Features. In IEEE Conference on Computer Vision and Pattern Recognition, 2001(1):511-518)
[20] Michael Jones, Paul Viola. Robust Real-time Object Detection.Cambridge Research Laboratory Technical Report Series. 2003
[21] Michael Jones, Paul Viola. Fast Multi-view Face Detection. IEEE Conference on Computer Vision and Pattern Recongnition(CVPR). 2003
[22] Rainer Lienhart, Alexander Kuranov, Vadim Pisarevsky, Empirical Analysis of Detection Cascades of Boosted Classifiers for Rapid Object Detection, MRL Tech Report, May 2002.nical
[23] Opencv. [Online]. Available: http://sourceforge.net/projects/opencvlibrary/
[24]田欣,基于不同色彩空间的肤色模型,西安科技学院学报,2001,21(4):369~372
[25]徐艳敏,张星明,刘晓红,苏曙,基于色彩聚类的肤色分割方法的研究,广州航海高等学校学报,2004,12(1):24~27
[26] Kanade T. Picture processing by computer complex and recognition of human faces [PH.D. Thesis]. Kyoto: Kyoto University, 1973.
[27] Brunelli R, Poggio T. Face recognition: Features versus templates. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1993, 15(10): 1042~1052
[28] Feng GC, Yuen PC. Variance projection function and its application to eye detection for human face recognition. Pattern Recognition Letters, 1998, 19(9): 899~ 906
[30] Smith S M, Brady J M. SUSAN - A new approach to low level image processing [J], Int Journal of Computer Vision, 1997.23(1): 126
[31]杨莉,张弘,李玉山,一种快速自适应RSUSAN角点检测算法[J],计算机科学,2004,31(5):198~200
[32] Jin Ren-jie, Ye Xiu-qing, Xu Sheng-rong, Computer image processing [M], Hangzhou: Zhejiang University Publishing Company, 1990
[2]李刚,高政、人脸自动识别方法综述、计算机应用研究、2003, Vol 8:4-9
[3] C.kotropoulos, I. Pitas. Rule-Based Face Detection in Frontal Views. Proc. Of Int '1 Conf. Acoustics, Speech and Signal Processing. 1997:25(4)
[4] G.Yang, T.S. Huang. Human Face Detection in Complex Background. Patten Recognition.1994, 27(l1):53-6
[5] D.Chetverikov, A.Lerch. Multiresolution Face Detection. Theoretical Foundations of Computer Vision. 1993, 69: 131-140
[6] H.P. Graf, T. Chen, E. Petajan, E. Cosatto. Locating Faces and Facial Parts. Proc. First Int’l Workshop Automatic Face and Gesture Recognition. 1995: 41-46
[7] T.K. Leung, M.C. Burl, P. Perona. Finding Faces in Cluttered Scenes Using Random Labeled Graph Matching. Proc. Fifth IEEE Int’l Conf. Computer Vision. 1995: 637-644
[8] K.C. Yow, R. Cipolla. A Probabilistic Framework for Perceptual Grouping of Features for Human Face Detection. Proc. Second Int’l Conf. Automatic Face and Gesture Recognition. 1996: 16-21
[9] Augusteijn MF,Skufca TL.Identification of human faces through texture-based feature recognition and neural network technology[C].Proc IEEE Int Joint Conf Neural Networks 1993:392–398.
[10] Y.Dai,Y.Nakano. Face-Texture Model Based on SGLD and Its Application in Face Detection in a Color Scene[C].Pattern Recognition.1996,V01.29,no.6:1007-1017.
[11] Yoo T W, Oh I S.A fast algorithm for tracking human faces based on chromatic histograms[C].Pattern Recognition Letters,1999,20(10):967-978.
[12] Moghaddam B,Pentland A.Probabilistic visual learning for object representation [C].IEEE Trans Pattern Analysis and Machine Intelligence,1997,19(7):696-710.
[13] Rowley H A,Baluja S,Kanade T.Rotation invariant neural network-based face detection[Z].Carnegie Mellon University,Pittsburgh PA:Technical Report CMU-CS-97-201,1997.
[14] Osuna E,Freund R,Girosi F.Training support vector machines:An application to face detection[C].In:Proc Computer Vision and Pattern Recognition,Puerto Rico,1997:130-136.
[15] Nefian A V,Hayes M H.Face detection and recognition using hidden Markov models[C].In:Proc IEEE Conference on Image Processing,Chicago,1998:141-145.
[16]张舒,褚艳利,赵开勇,GPU高性能运算之CUDA,中国水利水电出版社2009 36-58 120-140
[17] NVIDIA CUDA Best Practices Guide, NVIDIA,Version3.0 Apr. 2010.[Online].Available:http://cuda.csdn.net/download/CUDA Toolkit version 1.1 for Windows XP/C/doc/NVIDIA_CUDA_BestPracticesGuide.pdf
[18] Schapire R. E. A Brief Introduction of Boosting. Proceedings of the 16th International Joint Conference on Artificial Intelligence. 1999
[19] Viola P and Jones M. Rapid Object Detection using a Boosted Cascade of Simple Features. In IEEE Conference on Computer Vision and Pattern Recognition, 2001(1):511-518)
[20] Michael Jones, Paul Viola. Robust Real-time Object Detection.Cambridge Research Laboratory Technical Report Series. 2003
[21] Michael Jones, Paul Viola. Fast Multi-view Face Detection. IEEE Conference on Computer Vision and Pattern Recongnition(CVPR). 2003
[22] Rainer Lienhart, Alexander Kuranov, Vadim Pisarevsky, Empirical Analysis of Detection Cascades of Boosted Classifiers for Rapid Object Detection, MRL Tech Report, May 2002.nical
[23] Opencv. [Online]. Available: http://sourceforge.net/projects/opencvlibrary/
[24]田欣,基于不同色彩空间的肤色模型,西安科技学院学报,2001,21(4):369~372
[25]徐艳敏,张星明,刘晓红,苏曙,基于色彩聚类的肤色分割方法的研究,广州航海高等学校学报,2004,12(1):24~27
[26] Kanade T. Picture processing by computer complex and recognition of human faces [PH.D. Thesis]. Kyoto: Kyoto University, 1973.
[27] Brunelli R, Poggio T. Face recognition: Features versus templates. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1993, 15(10): 1042~1052
[28] Feng GC, Yuen PC. Variance projection function and its application to eye detection for human face recognition. Pattern Recognition Letters, 1998, 19(9): 899~ 906
[30] Smith S M, Brady J M. SUSAN - A new approach to low level image processing [J], Int Journal of Computer Vision, 1997.23(1): 126
[31]杨莉,张弘,李玉山,一种快速自适应RSUSAN角点检测算法[J],计算机科学,2004,31(5):198~200
[32] Jin Ren-jie, Ye Xiu-qing, Xu Sheng-rong, Computer image processing [M], Hangzhou: Zhejiang University Publishing Company, 1990