计算机应用 ›› 2015, Vol. 35 ›› Issue (7): 2062-2066.DOI: 10.11772/j.issn.1001-9081.2015.07.2062

• 虚拟现实与数字媒体 • 上一篇    下一篇

在线融合特征的眼睛状态识别算法

徐国庆1,2   

  1. 1. 武汉工程大学 计算机科学与工程学院, 武汉 430205;
    2. 武汉工程大学 智能机器人湖北省重点实验室, 武汉 430205
  • 收稿日期:2015-01-23 修回日期:2015-03-26 出版日期:2015-07-10 发布日期:2015-07-17
  • 通讯作者: 徐国庆(1974-),男,江苏徐州人,副教授,博士,主要研究方向:计算机视觉,xgqtiger@163.com
  • 基金资助:

    湖北省自然科学基金资助项目(2014CFB786);湖北省高等学校青年教师深入企业行动计划项目(XD2014146);武汉工程大学科学研究基金资助项目。

Eye state recognition algorithm based on online features

XU Guoqing1,2   

  1. 1. School of Computer Science and Engineering, Wuhan Institute of Technology, Wuhan Hubei 430205, China;
    2. Hubei Provincial Key Laboratory of Intelligent Robot (Wuhan Institute of Technology), Wuhan Hubei 430205, China
  • Received:2015-01-23 Revised:2015-03-26 Online:2015-07-10 Published:2015-07-17

摘要:

针对人脸视频中眼睛定位精度影响眼睛状态识别正确率问题,提出了一种融合在线肤色模型的眼睛状态识别算法。首先,在人脸主动表观模型(AAM)定位的基础上,使用当前用户的肤色特征,建立在线肤色模型;其次,在初步定位的眼睛区域,再次使用在线肤色模型,定位内外眼角点的精确位置,并利用眼角点的位置信息提取精确的眼睛区域;最后,提取眼睛区域的局部二值特征(LBP),使用支持向量机(SVM)算法,实现对眼睛睁闭状态的鲁棒识别。实验结果表明,对比全局定位的眼角点定位算法,该算法可以进一步降低眼角点的对齐误差,在低分辨人脸中使用在线融合特征的睁闭眼状态的准确识别率分别为95.03%及95.47%,分别比直接使用Haar特征和Gabor特征的识别率提升2.9%和4.8%,在实时人脸视频中,使用在线特征可以明显提高眼睛状态识别效果。

关键词: 人机交互, 肤色模型, 特征定位, 眼睛状态识别, 局部特征

Abstract:

Focusing on the issue that the eye localization accuracy drastically affects the correct recognition rate of the eye state, an eye state recognition algorithm combined with online skin feature model was proposed. Firstly, an online skin model was established by fusing the Active Appearance Model (AAM) of the received face image and the skin characteristics of the active user. Secondly, in the preliminary positioned eye area, the online skin model was used again to calculate the precise location of the inner and outer corners of the eyes, and the optimal eye positions were computed by reference of the eye corners. Finally, the Local Binary Pattern (LBP) in the eye area was extracted, and the close and open state of the eyes was recognized effectively based on the Support Vector Machine (SVM). In the comparison experiments with eye corners location algorithm of global localization, the location error was further reduced, and in a low resolution face image, the average recognition accuracy of open eye state and close eye state were 95.03% and 95.47% respectively. Compared with the algorithms based on Haar features and Gabor features, the efficiency increased by 2.9% and 4.8% respectively. The theoretical analysis and simulation results show that the algorithm based on online feature can effectively improve the recognition efficiency of the eye state from real-time face video.

Key words: human computer interaction, skin model, feature location, eye state recognition, local feature

中图分类号: