论文总字数：32565字

摘 要

手势识别在哑语识别、无人驾驶车辆指挥等领域中有着重要意义，现在已经在设备参数获取，信息传递，设备调节和设备控制方面拥有较高的自动化和智能化水平。在现有的监控系统中增强智能视频分析能力，提取手和手臂区域，有助于更准确判断手势，为特定的手势识别奠定了基础。实现视频中交通指挥手势识别系统的手及手臂检测、分割及提取。

目标检测和图像分割是对监控视频图像进行处理和分析得到的结果，该目标的实现将提高无人驾驶车辆的自动化和智能化程度，为真正无人驾驶创造条件。手部的有效分割能够提高手和手臂动作的识别率，在无人驾驶车辆上应用能够使交通更加安全便捷高效。图像的分割识别是先阶段研究的热点话题，对手的分割提取在现实中有非常广泛的应用，因此很有必要研究这一课题。

本文首先分析了图像分割的研究背景及意义，国内外研究现状。然后使用YOLO网络对人体进行检测，这是后面识别手臂一个预处理的部分。其次给出了一些传统的分割图像的算法并分析了其优缺点。接着分析了基于深度网络的手及手臂分割的一些算法，本文对目标检测和图像分割进行了说明，主要采用了deeplabv3的算法。在tensorflow框架下使用deeplabv3网络实现了图像分割，对其进行了实验验证，能够分割出手的区域且具有良好的准确性。并对这种算法进行结构分析与实验验证，最后提出了一些缺点和不足，总结全文并对未来展望。论文结尾的附录附有软件代码。

关键词：手和手臂区域提取 目标分割 卷积神经网络

Hand and arm region extraction for specific traffic command gesture recognition of driverless vehicles

ABSTRACT

Gesture recognition plays an important role in the fields of speech recognition, driverless vehicle command and so on. Now it has a high level of automation and intelligence in equipment parameter acquisition, information transmission, equipment adjustment and equipment control. In the existing monitoring system, the ability of intelligent video analysis is enhanced to extract hand and arm regions, which is helpful to judge gesture more accurately and lay a foundation for specific gesture recognition. To realize the hand and arm detection, segmentation and extraction of the traffic command gesture recognition system in video.

Target detection and image segmentation are the results of processing and analyzing surveillance video images. The realization of this target will improve the automation and intelligence of driverless vehicles, and create conditions for real driverless vehicles. The effective segmentation of hand can improve the recognition rate of hand and arm movements, and the application in driverless vehicles can make the traffic more safe, convenient and efficient. Image segmentation and recognition is a hot topic in the first stage of research. The segmentation and extraction of the opponent has a very wide range of applications in reality, so it is necessary to study this topic.

Firstly, this paper analyzes the research background and significance of image segmentation, and the research status at home and abroad. Then we use Yolo network to detect the human body, which is a preprocessing part of the back arm recognition. Secondly, some traditional image segmentation algorithms are given and their advantages and disadvantages are analyzed. Then, some algorithms of hand and arm segmentation based on depth network are analyzed. In this paper, the target detection and image segmentation are described, mainly using the deep labv3 algorithm. Under the framework of tensorflow, image segmentation is realized by using deep labv3 network, which is verified by experiments. It can segment the area of hand and has good accuracy. At the end of this paper, some shortcomings and deficiencies are put forward, and the full text is summarized. The appendix at the end of the paper is attached with software code.

Key words: hand and arm area extraction target segmentation convolution neural network

目 录

摘 要 I

Abstract II

目 录 III

第一章 绪论 1

1.1 概述及发展现状 1

1.2 国内外研究现状 2

1.3 本文内容安排 3

第二章 人体的提取及检测 4

2.1人体检测的方法 5

2.2基于神经网络的人体检测 6

2.2.1 YOLO网络原理及结构 6

2.2.2 YOLO的损失函数 7

2.3 实验结果与分析 7

第三章 手臂区域分割及其相关方法 9

3.1分割的基本概念 10

3.2传统的分割方法 10

3.2.1传统的分割方法——帧差法 10

3.2.2 传统的分割方法——基于肤色的手及手臂分割 10

3.3 基于深度学习的方法 11

3.3.1 FCN网络 13

3.3.2 DeepLab系列 14

3.4 deeplabv3原理及结构 14

3.5 deeplabv3的损失函数 17

第四章 基于deeplabv3网络的手及手臂区域提取 19

4.1 windows下框架的搭建及库安装 19

4.1.1 tensorflow环境介绍 19

4.1.2 框架搭建方法和对应库的安装 20

4.2 数据集的制作 22

4.3修改及训练数据 23

4.4测试 23

4.5 实验的结果与分析 24

第五章 论文总结与展望 27

5.1 论文总结 27

5.2 研究展望 27

参考文献 29

致 谢 31

请支付后下载全文，论文总字数：32565字