![](https://csdnimg.cn/release/download_crawler_static/87966027/bg1.jpg)
信息科学与技术学院毕业论文
课题名称: 基于特征识别的人脸检测系统
学 院: 信息科学与技术学院
完成日期: 二○一七年五月十九日
![](https://csdnimg.cn/release/download_crawler_static/87966027/bg2.jpg)
摘要
I
摘 要
我的毕业设计题目是基于特征的人脸检测系统,这个系统不仅仅能够检测人脸,
还具有识别人脸的功能。检测人脸检测部分的算法采用的是于仕祺老师的 LBP 特征
加 Gentle AdaBoost分类器相结合的算法,提取识别特征部分的算法采用的是 Google
在 2015 年提出的基于深度学习策略的一种人工神经网络 FaceNet,较为新颖,其准
确率高,在光照不足,姿态和表情变化剧烈时仍能保持稳定,具有很强的鲁棒性。
该系统的界面使用 MFC 编写,在具体实现中了应用了多线程编程技术实现了一个简
单的生产者消费者模型,从而提高了系统的识别效率,另外,对人脸的识别模块还
使用了 Python,C++混合编程技术引入了 Google 的开源深度学习框架 Tensorflow作
为对 FaceNet的具体实现,数据库使用的是 SQL Server2012,连接数据库使用的是微
软公司的 ADO 组件。该系统主要有信息采集模块和实时监控模块两个部分,前者完
成对任务样本的信息采集工作,后者完成在实时监控的情况下对出现在画面中的人
脸进行检测和识别,检测部分的速度可以达到 40~60 的 FPS,识别部分由于计算量较
大,只能达到 2~5 的 FPS。该系统经过简单的硬件支持和部署之后,基本可以完成在
实际场景中的简单应用,具有一定的学术研究和实际应用价值。
关 键 词:人脸检测;人脸识别;机器学习;Tensorflow;实时监控
![](https://csdnimg.cn/release/download_crawler_static/87966027/bg3.jpg)
ABSTRACT
II
ABSTRACT
The topic of this graduation project is Face Detection System based on characteristics which
achieves the face detection and face recognition two functions. The algorithm of face detection part
uses a kind of enhanced algorithm based on LBP feature and Gentle AdaBoost classifier proposed by
ShiQi Yu,the algorithm of extracting face feature used in recognition part uses a kind of manual
neural network FaceNet based on deep learning strategy proposed by Google in 2015.FaceNet has
reached high arruracy and it is robustness to the change of illumination,posture and expression.The
interface of this system if written in MFC,and in real implementation,the application of
multi-threaded programming technology realizes a simple producer and consumer model which
accelerate the whole recognition efficiency of the system,in addition,the recognition part also uses
the Python,C++ mixed programming technology which introduces Google’s open-source deep
learning framework Tensorflow as a concrete implementation of FaceNet,the database is using SQL
Server 2012,the link of database uses Microsoft’s ADO components.The system consists of two parts:
the information collection module and the real-time monitoring module,the former completes the
information collection of human samples,and the latter completes the detection and recognition of
the faces that appear in the picture in the case of real-time monitoring.The speed of detection part
can reaches to 40~60 FPS,the recognition part can only reach to 2~5 FPS, for the large calculation
cost.After some simple hardware support and deployment,the system can be used in actual scene
for simple application which has certain research and practical application value.
KEY WORDS :
Face detection; Face recognition; Machine learning; Tensorflow; Real-time
monitoring
![](https://csdnimg.cn/release/download_crawler_static/87966027/bg4.jpg)
石河子大学信息科学与技术学院毕业论文
1
目 录
1 绪论 .................................................................3
1.1课题 ............................................................................................................................................ 3
1.2课题背景..................................................................................................................................... 3
1.3课题研究目的及意义 ................................................................................................................. 3
1.3.1 研究目的 ........................................................................................................................ 3
1.3.2 研究意义 ........................................................................................................................ 4
1.4国内外研究现状 ......................................................................................................................... 5
1.4.1 国外 ................................................................................................................................ 5
1.4.2 国内 ................................................................................................................................ 6
1.5设计时间..................................................................................................................................... 7
1.6内容及分工 ................................................................................................................................. 7
1.6.1 内容 ................................................................................................................................ 7
1.6.2 成果 ................................................................................................................................ 7
2 理论和技术 ...........................................................8
2.1理论 ............................................................................................................................................ 8
2.1.1 检测部分的 LBP 特征+Gentle AdaBoost 分类器 ....................................................... 8
2.1.2 识别部分的 Google FaceNet ..................................................................................... 11
2.2技术 .......................................................................................................................................... 15
2.2.1 MFC 简介 ...................................................................................................................... 15
2.2.2 Tensorflow 简介 ........................................................................................................ 17
2.2.3 ADO 组件简介 .............................................................................................................. 18
3 需求分析及概要设计 ..................................................19
3.1 需求分析.................................................................................................................................. 19
3.2 概要设计.................................................................................................................................. 19
4 实 现 ..............................................................20
4.1核心理论的实现 ....................................................................................................................... 20
4.1.1检测部分的实现 .................................................................................................................... 20
4.1.2提取识别特征部分的实现 .................................................................................................... 21
4.2类 .............................................................................................................................................. 24
4.2.2 TensorflowInference 类 .......................................................................................... 26
4.2.1 ADOConn 类 .................................................................................................................. 28
4.3主界面....................................................................................................................................... 29
4.3.1 流程............................................................................................................................... 29
4.3.2 实现............................................................................................................................... 30
4.4信息采集................................................................................................................................... 32
4.4.1 流程 .............................................................................................................................. 32
4.4.2 实现 .............................................................................................................................. 32
4.5实时监控................................................................................................................................... 34
4.5.1 流程 .............................................................................................................................. 35
4.5.2 实现 .............................................................................................................................. 36
4.6数据库....................................................................................................................................... 38
5 测试 ................................................................40
5.1 主界面测试 ...................................................................................................................... 40
5.2 信息采集测试 .................................................................................................................. 41
5.3 实时监控测试 .................................................................................................................. 41
![](https://csdnimg.cn/release/download_crawler_static/87966027/bg5.jpg)
石河子大学信息科学与技术学院毕业论文
2
6 结论 ................................................................43
7 致 谢 ..............................................................44
8 参考文献 ............................................................45