[1]龚格格,吴 珊,郭湘南.人脸检测算法的优化[J].计算机技术与发展,2019,29(06):47-51.[doi:10. 3969 / j. issn. 1673-629X. 2019. 06. 010]
　GONG Ge-ge,WU Shan,GUO Xiang-nan.Optimization of Face Detection Algorithm[J].,2019,29(06):47-51.[doi:10. 3969 / j. issn. 1673-629X. 2019. 06. 010]
点击复制

人脸检测算法的优化()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
29
期数:
2019年06期
页码:
47-51
栏目:
智能、算法、系统工程
出版日期:
2019-06-10

文章信息/Info

Title:
Optimization of Face Detection Algorithm
文章编号:
1673-629X(2019)06-0047-05
作者:
龚格格1 ; 吴 珊1 ; 郭湘南2
1. 武汉邮电科学研究院,湖北 武汉 430000;2. 武汉烽视威科技有限公司,湖北 武汉 430000
Author(s):
GONG Ge-ge1 ; WU Shan1 ; GUO Xiang-nan2
1. Wuhan Institute of Posts and Telecommunications Science,Wuhan 430000,China;2. Wuhan Feng Visual Technology Co. ,Ltd. ,Wuhan 430000,China
关键词:
人脸检测; 候选框生成算法; Faster RCNN; 快速级联卷积神经网络模型; 网络性能
Keywords:
face detection;  candidate frame generation algorithm;  Faster RCNN;  fast cascade convolution neural network model; network performance
分类号:
TP301.6
DOI:
10. 3969 / j. issn. 1673-629X. 2019. 06. 010
摘要:
面部特征被广泛应用于一系列视频监控系统,其中公安系统中人脸检测模块尤为突出。 由于人脸的巨大视觉变化,如遮挡、光照、大的姿态变化问题使人脸检测一直存在着瓶颈,在实际应用中这些问题依旧很常见。 对此,文中通过简要介绍候选框生成算法,同时结合 Faster RCNN、联合人脸检测和对齐的级联卷积神经网络框架的优缺点进行分析和改进,提出了快速级联卷积神经网络模型。 由于候选框网络和 RoI 检测网络共享卷积层,在候选框网络中使用多层卷积层信息,采用 RoI 池化和 L2 归一化将身体信息与面部信息进行融合,实现结合身体上下文信息来处理较小的人脸区域,并对数据集进行测试来验证模型的有效性,弥补因视觉变化导致人脸检测中的不足,提高人脸检测网络性能。
Abstract:
Facial features are widely used in a series of video monitoring systems,among which the face detection module in the public security system is particularly prominent. Due to the huge visual changes of face,such as occlusion,illumination and large posture changes,human detection has always been a bottleneck,and these problems are still very common in practical applications. For this, through the brief introduction of the candidate generation algorithm,at the same time, combined with Faster RCNN, analysis and improvement of the advantages and disadvantages of face detection and alignment cascaded convolutional neural network framework,we propose a fast cascaded convolution neural network model. Since the candidate box and RoI detection networks share the convolution layer,multi-layer convolutional layer information is used in the candidate box network. The RoI pooling and L2 normalized body and facial information are used to fuse,dealing with the smaller face region with the physical context information,and testing data sets to verify the validity of the model,which makes up for the deficiency of face detection caused by visual changes and improves the performance of face detection network.

相似文献/References:

[1]李宏伟 陶亮.复杂背景不同光照条件下彩色图像中人脸检测[J].计算机技术与发展,2009,(02):52.
　LI Hong-wei,TAO Liang.Face Detection in Complicated Backgrounds and Different Illumination Conditions of Color Images[J].,2009,(06):52.
[2]俞扬信 严云洋.视频序列中的人脸检测与定位算法研究[J].计算机技术与发展,2009,(02):109.
　YU Yang-xin,YAN Yun-yang.Algorithm Study of Face Detection and Location in Video Sequence[J].,2009,(06):109.
[3]金燕 陶亮.基于人眼定位的人脸检测与归一化算法[J].计算机技术与发展,2009,(04):95.
　JIN Yan,TAO Liang.Face Detection and Normalization Based on Localization of Human Eyes[J].,2009,(06):95.
[4]佘九华 王敬东 李鹏.基于Camshift的人脸跟踪算法[J].计算机技术与发展,2008,(09):12.
　SHE Jiu-hua,WANG Jing-dong,LI Peng.Face Tracking Algorithm Based on Camshift[J].,2008,(06):12.
[5]李启娟 李金屏.基于轮廓信息的人脸检测[J].计算机技术与发展,2008,(09):108.
　LI Qi-juan,LI Jin-ping.Face Detection Based on Contour Information[J].,2008,(06):108.
[6]陈伟琦 梁一川 易强 秦文虎.基于肤色和Adaboost算法的人脸检测研究[J].计算机技术与发展,2008,(12):44.
　CHEN Wei-qi,LIANG Yi-chuan,YI Qiang,et al.Face Detection Based on Skin Color and Adaboost Arithmetic[J].,2008,(06):44.
[7]贾灵芝 李岚 钱坤喜.基于自适应光线补偿的人脸检测算法[J].计算机技术与发展,2008,(12):120.
　JIA Ling-zhi,LI Lan,QIAN Kun-xi.Face Detection Algorithm Based on Self- adaptive Light Compensation[J].,2008,(06):120.
[8]袁芬萍 季桂树.基于SVM的三阶段人脸检测方法的研究与应用[J].计算机技术与发展,2007,(09):133.
　YUAN Fen-ping,JI Gui-shu.Research and Application about Three- Stage Face Detection Method Based on SVM[J].,2007,(06):133.
[9]王晶 杨煜.基于边缘方向直方图的Adaboost人脸检测[J].计算机技术与发展,2007,(12):5.
　WANG Jing,YANG Yu.Adaboost for Face Detection Based on Edge Orientation Histograms[J].,2007,(06):5.
[10]邵平 杨路明 曾耀荣.计算旋转Harr型特征的积分图像算法改进[J].计算机技术与发展,2006,(11):146.
　SHAO Ping,YANG. Lu-ming,ZENG Yao-rong.An Improved Algorithm of Integral Image for Computing Rotated Harr- Like Features[J].,2006,(06):146.

更新日期/Last Update: 2019-06-10