基于 AdaBoost 和 Kalman 算法的人眼检测与跟踪*

南京大学学报(自然科学版) ›› 2010, Vol. 46 ›› Issue (6): 681–687.

基于 AdaBoost 和 Kalman 算法的人眼检测与跟踪*

严超 1 , 王元庆 1 ** , 张兆扬 2

出版日期:2015-04-03 发布日期:2015-04-03
作者简介: (1. 南京大学电子科学与工程系, 南京, 210093; 2. 上海大学新型显示技术及应用集成教育部重点实验室, 上海, 200444)
基金资助:
国家自然科学基金( 608320036) , 新型显示技术及应用集成教育部重点实验室基金( P200902)

Eye detection and tracking based on AdaBoost and Kalman algorithms

Yan Chao 1 , Wang Yuan Qing 1 , Zhang Zhao Yang 2

Online:2015-04-03 Published:2015-04-03
About author: ( 1. Department of Electronic Science and Engineering, Nanjing University, Nanjing, 210093, China;
2. Key Laboratory of Advanced Display and System Application, Shanghai University, Ministry of Education, Shanghai, 200444, China)

摘要/Abstract

摘要： 自由立体显示装置中, 人眼位置的探测与跟踪是关键技术之一, 由于其具有的学术及社会价值, 人眼检测近年来成为模式识别领域中的一个研究热点. 人眼检测要求准确、实时, 为了满足上述技术
需求, 提出了一种基于 AdaBoost 和 Kalman 算法的人眼检测新方法. 新方法使用人脸人眼!的两步检测模式. 首先利用连续 AdaBoost 算法遍历图像, 完成人脸定位和人眼的初定位工作, 并在人眼定位位
置进行标定; 然后进一步利用连续 AdaBoost 算法在上述标定位置附近完成人眼的精定位工作, 最后, 利用 Kalman 算法对已经检测到的人眼位置进行跟踪, 以提供下一帧图像中人眼可能存在的区域, 使得在
下一帧图像中对这些区域优先进行检测.在实验条件为 Windows XP, PentiumIV, 512Memory, 2-4GHz 的情况下, 对于 640∀ 480 分辨率的连续视频, 系统的人眼检测率达到 93 ? 5%; 人眼检测平均时间小于 10 ms/帧, 达到了实时性的要求. 同
时对于人脸表情变化和人脸小角度倾斜也具有鲁棒性.

Abstract: Locating and tracking the human eyes is one of the critical technologies in free stereoscopic display system. Owing to its high academic and social value, eye detection has been becoming a hotspot in pattern
recognition recently. Eye detection should be accurate and real -time. To satisfy these technical needs, a new eye detection method based on AdaBoost and Kalman algorithms is discussed and tested. The new method includes two
parts: face detection and eye detection. First, it utilizes the real AdaBoost to traverse the whole image, locating human faces and completing the preliminary work of human eyes location, and demarcate the positions of human
eyes; then it employs the real AdaBoost to locate human eyes precisely near the positions demarcated; finally, it uses the Kalman algorithm to trace the human eyes, which have been detected, to provide the potential human eyes#
area in the next frame, and these area should be detected first in the next frame. Under the circumstances of Windows XP, PentiumIV, 512Memory, 2 -4 GHz, for a video sequence of 640 ∀
480pixel images, eye detection rate of the new method is 93 .5%; the average processing time for each image is less than 10 ms, satisfying the need of real ?time; this new method is also robust when there is variation of facial
expression or a little degree leaning of human face.

严超 1 , 王元庆 1 ** , 张兆扬 2. 基于 AdaBoost 和 Kalman 算法的人眼检测与跟踪*

[J]. 南京大学学报(自然科学版), 2010, 46(6): 681–687.

Yan Chao 1 , Wang Yuan Qing 1 , Zhang Zhao Yang 2
. Eye detection and tracking based on AdaBoost and Kalman algorithms
[J]. Journal of Nanjing University(Natural Sciences), 2010, 46(6): 681–687.

参考文献

[ 1 ] Horng W B, Chen C Y. Driver fatigue detection based on eye tracking and dynamic template matching. Proceedings of IEEE International Conference on Networking, Sensing and Contro1. Taipei. China: IEEE Press. 2004: 7~ 12.
[ 2 ] Zhang J, Yang X F, Zhao M L. Eyes detection based on Hough trans form. Compute Engineering and Application, 2005, 27: 43~ 44.
[ 3 ] Yano K, Lshihara K, Btakikawa, et al. Detection of eye blinking from video camera with dynamic ROI fixation. IEEE International Conference on Systems, Man, and Cybernetics, 1999, 6: 335~ 339.
[ 4 ] Mao M F, Zhang X M, Guo Y C. Fast rule based eyes location algorithm. Computer Engineering, 2004, 30(9): 5.
[ 5 ] Viola P, Jones M. Rapid object detection using a boosted cascade of simple features. Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, Kauai Hawaii, USA, 2001, 905~ 910.
[ 6 ] Zhao N. Face detection based on AdaBoost. Bachelor T hesis, Peiking University, 2005.
[ 7 ] Freund Y, Schapire R E. A decision-theoretic generalization of on?line learning and an application to boosting. Journal of Computer and System Sciences, 1997, 55( 1) : 119~ 139.
[ 8 ] Schapire R B, Singer Y. Improved boosting algorithms using confidence ?rated predictions. Machine Learning, 1999, 37(3): 297~ 336.
[ 9 ] Vapnik V. The nature of statistical learning theory. New York: Springer Verlag, 1995, 5~ 13.
[ 10] Freund Y, Schapire R E. Experiments with a new boosting algorithm. Proceedings of The 13 th Conference Machine Learning. SanFrancisco: M organ Kaufmann, 1996, 148~ 156.
[ 11] Zhao Y, Liu H X, Wang Z Y, et al. An improved nearest neighbor searching method for classification problems. Journal of Nanjing University( Natural Sciences) , 2009, 45(4): 455~ 462.
(肇莹, 刘红星, 军王仲宇等. 最近邻搜索用于分类问题的一种改进. 南京大学学报( 自然科学), 2009, 45(4): 455~ 462) .
[ 12] Shi Z K. Calculation method of optical estimation. Berjing: Science Press, 2001.
[ 13] Welch G. An introduction to the kalman filter. http: / /info. acre. org/pubs/toc/cRn0tice. html, 2001.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed