《计算机应用研究》|Application Research of Computers

一种自下而上的人脸检测算法

Bottom-up face detection algorithm

免费全文下载 (已被下载 次)  
获取PDF全文
作者 张宁,伍萍辉
机构 河北工业大学 电子信息工程学院 天津市电子材料与器件重点实验室,天津 300401
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2019)06-062-1897-04
DOI 10.19734/j.issn.1001-3695.2018.01.0032
摘要 针对在非控条件下的人脸检测经常遇到的问题,如复杂的人脸姿态表情、严重的人脸遮挡、外界环境背景复杂、光照条件差、小脸等提出了一种自下而上的人脸检测方法。自下而上的人脸检测是基于深度学习的,先进行人脸相关关键点检测和关键点之间的位置关系检测再进行人脸检测。网络结构采用稠密网络进行图像特征提取,提取到的特征传送给6个级联网络,每个级联网络由两个分支网络构成,分支网络1用来预测人脸相关关键点位置坐标,分支网络2用来预测关键点之间的位置关系。利用得到的关键点位置和位置关系进行人脸检测。在FDDB测试集上进行了验证,取得了0.98的成绩,并可以在输入图像分辨率为1 920×1 080的情况下,能检测到的最小人脸分辨率为10×10,使用GPU Nvidia GeForce GTX 1070最快能达到17 fps。
关键词 人脸检测; 深度学习; 关键点检测; 自下而上
基金项目
本文URL http://www.arocmag.com/article/01-2019-06-062.html
英文标题 Bottom-up face detection algorithm
作者英文名 Zhang Ning, Wu Pinghui
机构英文名 Tianjin Key Laboratory of Electronic Materials & Devices,School of Electronic & Information Engineering,Hebei University of Technology,Tianjin 300401,China
英文摘要 Faced with the problems often encountered in face detection under non-controlled conditions, such as complex facial expression, serious face occlusion, complex external environment, poor lighting conditions, tiny face, etc, this paper proposed a bottom-up face detection method. Bottom-up face detection was based on deep learning, face detection and key points of the first position-related key detection and then face detection. Convolution neural network structure using dense network for image feature extraction, it transmitted the extracted features to 6 cascaded networks, each of which consisted of two branch networks. It used branch network 1 to predict the coordinates of face-related key points, and used branch network 2 to predict the position between key points relationship. This paper performed face detection by using the obtained key point position and position relationship. It verifies the FDDB test set and achieves 0.98 results, and the smallest face resolution 10×10 can be detected at the resolution of input image 1920×1080, used the GPU Nvidia GeForce GTX 1070 for up to 17 fps video detection.
英文关键词 face detection; deep learning; key point detection; bottom-up
参考文献 查看稿件参考文献
 
收稿日期 2018/1/18
修回日期 2018/3/15
页码 1897-1900,1906
中图分类号 TP391.41;TP301.6
文献标志码 A