ISSN: 1673-825X    Imprint: Chongqing University of Posts and Telecommunications Journal
Volume listings
  • Current volume
  •  
  • Search by keywords:
  • Search by 
  • Keywords
  • To
  • Journal archive
Related Links
基于回归的人脸检测加速算法
A face detection acceleration algorithm based on regression
DOI:10.3979/j.issn.1673-825X.2019.04.016
Received:April 19, 2019  Revised:May 30, 2019
中文关键词:人脸检测  回归算法  人脸位置预测  视频帧
英文关键词:face recognition  regression algorithm  face position prediction  video frame
基金项目:吉林省省级科技创新专项资金(20190302026GX) ; 吉林省发改委产业技术研究与开发项目(2019C054-4);应用光学国家重点实验室开放基金(20173660)
Author NameAffiliationE-mail
WANG Dan Editorial Board of Journal of Bionic Engineering, Jilin University, Changchun 130022, P. R. China d_wang@jlu.edu.cn 
ZHAO Hongwei College of Computer Science and Technology, Jilin University, Changchun 130012, P. R. China zhaohw@jlu.edu.cn 
DAI Yi PetroChina Fushun Petrochemical Company, Fushun 113006, P. R. China  
WU Bin College of Computer Science and Technology, Jilin University, Changchun 130012, P. R. China  
Hits: 27
Download times: 15
中文摘要:
      为了提高视频中人脸检测的检测速度,采用回归分析方法预测连续视频中人脸中心位置坐标,并通过调整区域宽度系数确定人脸区域位置,从而提出一种人脸检测加速算法。该算法的主体框架采用VJ(Viola-Joines)结构,在人脸检测过程中,通过聚合通道特征和弱级联分类建立多尺度精细采样图像特征金字塔,并利用回归分析方法进行人脸中心位置坐标拟合,再采用粗粒度预测方法降低算法时间复杂度,最后通过优化人脸区域位置系数提高人脸检测准确率。在此基础上,又通过目标预测、跟踪算法进行人脸检测的二次加速。实验结果表明,该算法有效减少了视频人脸检测遍历区域,提高了人脸检测的检测速度,缩短了提取视频人脸特征区域的时间,更加适合视频人脸检测的实时性应用。
英文摘要:
      To increase the speed of face detection in the video, a regression analysis method is used to predict the coordinates of center of face in continuous videos, the region of face is determined by adjusting the region width coefficient, and then an acceleration algorithm for face detection is proposed. The VR framework is adopted in the algorithm. During the face detection, a multi-scale image feature pyramid of fine sampling is constructed using aggregation channel features and week classifier. The regression analysis method is used to fit the coordinates of the face center. The coarse-grained prediction method is used to reduce the time complexity of the algorithm. The accuracy of face detection is improved by optimizing the face width coefficient. Then, secondary acceleration of face detection is realized using object prediction and tracking algorithm. The results show this algorithm can reduce the traversal region of video face detection, increase the speed of face detection, and decrease the time cost for the extraction of video face feature region, which is more suitable for the real time application of video face detection.
HTML    View PDF   View/Add Comment  Download reader