摘 要

作为成长于21世纪的当代大学生，我们与机器人技术一同茁壮成长。仅仅是十几年的时间，我们见证了机器视觉的蓬勃发展。在日常生活中也许我们很少接触智能机器人，但是机器视觉的应用确是渗透进我们的生活的点点滴滴。与我们关系最紧密的是智能手机的发展，曾经手机解锁都是数字密码或者图案密码，渐渐演变为指纹识别。近几年，人脸识别解锁成为了很多智能手机的选择。这证明了机器视觉算法的精确性、实时性，特别是安全性有了高水平的保障。机器视觉的研究价值与市场前景一片光明。

最开始的机器人其实只是模仿人的动作的“机器”，而随着人们对人机交互和人工智能的重视，机器人变得真的像“人”一样能够“看”这个世界，“听”万物之声，还能像人一样进行“学习”。在人类引以为傲的围棋领域，这一象征着人类高水平智慧的游戏中，人工智能机器人击败了世界排名第一的柯洁。这一事件将人工智能机器人又一次推向了热议话题中心。当然人工智能机器人不光在智能上有所追求，事实上在科研领域如何使它更好的应用于服务社会大众的各行各业才是更多人追求的目标。

本文基于养老院这样的公共环境对机器视觉进行研究 。基于老年人这一特殊群体，我选择了行人检测和手势识别进行了深入学习和讨论。在行人检测方面提出了机器学习中的支持向量机（SVM）算法并结合方向梯度直方图（HOG）的算法，在手势识别中提出一种结合具有平移，旋转，缩放不变性的Hu矩和特殊颜色空间分割肤色的手势识别方法。所有以上的识别方法都要用到图像预处理，因此，本文花了一章来研究各种图像处理方式。

全文的主要创新点是给出了SVM HOG的行人检测算法，在手势识别中使用了一种兼容YCbCr和HSV颜色空间的自适应肤色分割模型。

关键词：行人检测；手势识别；SVM；HOG；颜色空间

Abstract

In this paper, As a contemporary college student who grew up in the 21st century, we thrive with robotics.For more than a decade, we have witnessed the flourishing of machine vision. We may not be exposed to intelligent robots in our daily lives, but the applications of machine vision are the drips that permeate our lives. The closest thing we have to our smart phones is the development of smart phones, which used to be digital or graphic codes that evolved into fingerprint recognition. In recent years, facial recognition has become a choice for many smart phones. This proves the accuracy and real time of the machine vision algorithm, especially the security with high level of security. The value of machine vision and the prospect of the market are bright.

At the beginning of the robot is just imitate human action "machine", and with the people to the attention of the human-computer interaction and artificial intelligence, robots become really like a "person" to "see" the world, "hear" the sound of all things, can also like people to "learn". In the field of go, which humans are proud to be, this symbol of the high level of human intelligence in the game of human intelligence, the artificial intelligence robot beats the world number one Kejie. The incident has pushed artificial intelligence into the center of the debate. The robot of artificial intelligence, of course, not only in seeking something smart, in fact in the field of scientific research how to make it better applied to service the public is more and more people from all walks of life the pursuit of goals.

This paper studies machine vision based on public environment such as nursing home. Based on this particular group of elderly people, I have chosen pedestrian detection and gesture recognition for further study and discussion. On pedestrian detection in the proposed machine learning support vector machine (SVM) algorithm and combined with the algorithm of gradient direction histogram (HOG), in the gesture recognition in this paper, a combined with translation, rotation and scaling invariance of Hu moment and special color space skin color segmentation method of gesture recognition. All of the above methods use image preprocessing, so this article has taken a chapter on various image processing methods.

The full text of the main innovation points are SVM HOG pedestrian detection algorithm, is given in the gesture recognition using a compatible YCbCr and adaptive skin color segmentation model of HSV color space.

Key Words: pedestrian detection；gesture recognition；SVM；HOG；color space

目 录

摘 要 I

Abstract II

第1章 绪论 1

1.1 研究背景 1

1.2 国内外发展历程与现状 1

1.3 研究目的与意义 2

1.4 本文介绍 3

第2章 图像处理 5

2.1 OpenCV简介 5

2.2 数字图像 5

2.2.1 HSV颜色空间 6

2.2.2 YCbCr颜色空间 9

2.2.3 颜色空间转换 10

2.3 图像预处理 11

2.4 canny边缘检测算法 12

第3章 手势识别 14

3.1 轮廓提取 14

3.2 轮廓的匹配——Hu不变矩 15

第4章 基于SVM和HOG的行人检测 17

4.1 支持向量机SVM 17

4.1.1 拉格朗日乘子法 18

4.1.2 KKT条件 18

4.1.3 核函数 19

4.2 HOG特征 20

第5章 行人检测与手势识别的实现 22

5.1 实验平台与流程 22

5.2 程序详细设计 23

5.2.1自适应肤色分割模型 24

5.2.2特征提取 25

5.3实验结果与分析 26

第6章 总结 29

参考文献 30

致谢 31

第1章 绪论

1.1 研究背景

当下的数字世界图像形式和视频形式的信息无所不在。而一直以来，计算机总是被动地接受键盘和鼠标等外部设备的输入信息，无法主动的从外部获取信息，并且主动的处理。计算机视觉的目的就是让计算机像人眼一样获取图像或视频信息，主动地从这个世界寻觅信息。