«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

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

卷:

31

期数:

2021年07期

页码:

134-139

栏目:

应用前沿与综合

出版日期:

2021-07-10

文章信息/Info

Title:

Face Detection in Advertising Recommendation System Based on Face-eyes Co-detector

文章编号:

1673-629X(2021)07-0134-06

作者:

苏 彬¹; 2; 梁 栋¹; 2

1. 南京航空航天大学 计算机科学与技术学院 模式分析与机器智能工业和信息化部重点实验室,江苏 南京 210016;

2. 软件新技术与产业化协同创新中心,江苏 南京 210016

Author(s):

SU Bin¹; 2; LIANG Dong¹; 2

1. MIIT Key Laboratory of Pattern Analysis and Machine Intelligence,School of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics, Nanjing 210016,China;

2. Collaborative Innovation Center of Novel Software Technology and Industrialization, Nanjing 210016,China

关键词:

人脸检测; 眼睛检测; 协同检测器; 优化 SSD 算法; 广告推荐; 目标人脸

Keywords:

face detection; eyes detection; co-detector; optimized SSD; advertising recommendation; key face

分类号:

TP39

DOI:

10. 3969 / j. issn. 1673-629X. 2021. 07. 022

摘要:

在使用人脸检测识别的广告推荐场景中目标用户是正在注视广告屏幕的用户,因此在做广告推荐时需要在检测人脸的同时检测两只眼睛,确保检测抓取的人脸是正对广告设备的人脸。 现有的人脸检测加人脸姿态评估算法对硬件资源消耗过大,在低成本硬件上无法保证业务的运行效果,所以该文首先对主流检测器进行全面的比较分析,然后提出了一种基于优化后 SSD 算法的人脸和双眼协同检测器,可以同时检测人脸和眼睛,并根据人脸和眼睛的位置判断它们是否属于广告推荐对象。 该方法增加了人脸和眼睛的位置信息,增加了特征图的大小,并增加了三个卷积层,以获得小目标的低层特征。 预测层的数量也被扩展以增加预测的可能性。 同时,该方法在实验比较和真实场景试验中均显示出良好的效果。

Abstract:

:In the scene of advertising recommendation using face detection and recognition,the target user is the user who is staring at theadvertising screen. Therefore,when making advertising recommendation,it is necessary to detect both eyes while detecting the face to ensure that the face captured by detection is facing the face of the advertising device. Existing face detection and face pose evaluation algorithms consume too much hardware resources and cannot guarantee the operation effect of services on low-cost hardware. Therefore,we first carry out a comprehensive comparative analysis of the mainstream detector,and then propose a face-eyes co-detector based on optimized SSD algorithm,which can simultaneously detect faces and eyes and judge whether faces and eyes belong to the recommended objects of advertising according to their positions. In this method,the position information of face and eyes is increased,the size of the feature map is increased, and three convolution layers are added to obtain the low - level features of small targets. The number of prediction layers has also been extended to increase the likelihood of prediction. Also the proposed method shows excellent results in both experimental comparison and the trials of the real scenes.

常用功能

更新日期/Last Update: 2021-07-10