论文总字数：31353字

摘 要

人物行为动作检测以及人体关键点识别是人工智能领域的热门方向之一，人体关键点检测对异常行为检测，动作分析，无人驾驶都具有基础性作用。

在本次毕业设计中，作者将人脸识别，年龄性别预测与动作识别进行了集成，最大程度上从图像里获取信息，除此以外本项目的动作数据库是自由可添加，比起一般的动作识别，能够更好适应日常拍摄图片，为了适应图片变化，还增加了图片补全的功能。

在开发语言上，本项目采用python作为开发语言，python语言具有简洁，高效，易于编写的特点。开发工具为Eclipse,Eclipse作为经典的Java开发工具，在安装pydev插件后也可以编写python代码，友好的编程界面易于程序员检查出错点修改错误。

本次毕业设计主要功能如下，对于给定视频中人物，第一步识别人物的年龄和性别，对人物信息进行初步判断，第二步对人物人脸定位并进行人脸识别，与数据库中已经认识的人脸进行比对，在一定阈值下寻找有无此人脸，若有则认为已识别，否则认为未识别。第三步对人物人体框架进行分析，运用神经网络前向传播得出人体关键点坐标，再将坐标提取出来，与数据库中已知的动作的坐标进行比对，最终得出动作。最后基于动作年龄性别对行为人的行为趋势进行预测，最终完成人物行为趋势的分析。

关键词：人脸识别，人体关键点检测，人体动作识别，年龄性别预测

Design and Implementation of Person Behavior Detection in Video Data

Abstract

Human key point detection is one of the hottest directions in the field of artificial intelligence. Human key point detection plays a fundamental role in abnormal behavior detection, motion analysis and unmanned driving.

In this graduation project, the author integrates face recognition, age and gender prediction and motion recognition to get information from images to the greatest extent. Besides, the action database of this project is free to add. Compared with general motion recognition, it can better adapt to daily photographs. In order to adapt to the changes of pictures, it also adds the function of picture completion.

In the development language, this project uses Python as the development language. The Python language is concise, efficient and easy to write. The development tool is Eclipse. Eclipse, as a classic Java development tool, can also write Python code after installing pydev plug-in. The friendly programming interface is easy for programmers to check errors and modify errors.

The main functions of this graduation project are as follows. For a given person in the video, the first step is to identify the age and gender of the person, to make a preliminary judgment on the information of the person, the second step is to locate and recognize the face of the person, to compare with the face already known in the database, to find out whether there is such a face under a certain threshold, and if there is one, to think that it has been recognized, otherwise it is not recognized. The third step is to analyze the human body frame, use the forward propagation of neural network to get the coordinates of key points of human body, then extract the coordinates, and compare with the coordinates of known actions in the database, and finally get the actions. Finally, based on the age and gender of the action, the behavior trend of the actor is predicted, and finally the analysis of the behavior trend of the character is completed.

Key Words：Face Recognition, Human Key Point Detection, Human Motion Recognition, Age and Gender Prediction

目录

摘 要 II

Abstract III

目录 IV

第一章 引言 1

1.1 系统开发的背景与意义 1

1.2 国内研究 2

1.3 国外研究 2

1.4 论文概述 3

第二章 本项目所用技术简单介绍 4

2.1神经网络 4

2.1.1前项传播 4

2.1.2反向传播 5

2.2卷积神经网络 7

2.3tensorflow 8

2.4 项目所用到的python工具 9

2.4.1matplotlib 9

2.4.2opencv 9

2.4.3numpy 9

2.4.4keras 9

2.5 face_recognition介绍 9

2.6 tf_pose_estimation介绍 10

第三章 人脸识别 10

3.1 人脸识别需求分析 10

3.2 人脸识别的实现 11

3.2.1 face_recognition的下载与安装 11

3.2.2 人脸数据库建立 12

3.2.3 人脸比对识别 13

第四章 年龄性别预测 17

4.1 年龄性别预测的需求分析 17

4.2 年龄性别预测的实现 17

4.2.1 age-gender-estimation 17

4.2.2 性别年龄预测代码细节 18

第五章 动作识别 20

5.1 动作识别需求分析 21

5.2 动作识别具体实现 22

5.3 项目优化与改进 31

第六章 总结展望 37

6.1 总结 37

6.2 展望 38

参考文献 39

致谢 42

第一章 引言

1.1系统开发的意义

请支付后下载全文，论文总字数：31353字