论文总字数：23552字

摘 要

近年来，经济的发展带来生活的不断进步，但是也引发了一些交通问题，如交通堵塞、交通事故等，并且这些问题有着逐步恶化的趋势，为了解决这些问题，基于视频的行人检测技术迫在眉睫。庆幸的是随着人工智能领域的兴起，越来越多的计算机领域技术被大力发展，行人检测技术也不例外。该技术就是通过公路上交通要道的视频探测器监视路面状况，并且能够在道路发生紧急状况比如车祸发生的时候及时地去通知警方进行紧急处理，而且可以保存当时的记录方便事后进行对事故的发生进行还原调查，该技术的用途之大已经被广泛应用于各行各业当中。

本论文主要通过以下方面展开研究。首先，介绍行人检测的背景，然后对传统的行人检测技术进行调查研究，发现以HOG SVM算法来实现传统的行人检测技术效果较好。然后在基于深度学习的行人检测算法中，研究者们发现卷积神经网络在检测识别类的问题上表现出出色的性能，检测率也比传统检测方式提高不少，所以颇受目前学习者们的青睐。由于YOLO v3是一个可以一次性地去预测多个位置和类别的卷积神经网络而且它的实时性较好，更容易针对视频的行人检测，因此最终实现了基于YOLO v3的行人检测技术，然后在现有的数据集上测试，通过对比传统的行人检测技术与基于深度学习的行人检测技术，验证各系统进行行人检测的实时性与准确率。

关键词： 行人检测 特征提取 深度学习 YOLO v3

Abstract

In recent years, the development of economy has brought about continuous progress of life, but also caused some traffic problems, such as traffic jams, traffic accidents, etc., and these problems have a trend of gradual deterioration. It is very vital to address these issues. Fortunately, with the rise of artificial intelligence, more and more computer technology is vigorously developed in the field, and pedestrian detection technology is no exception. The technology is through the highway traffic arteries video detector monitoring road conditions, and can occur at road emergencies such as car accident happens in a timely manner to inform the police for emergency treatment, but also can save these records in order to make survey . Meanwhile, this technology has been openly adapted to all aspects.

This paper studies pedestrian detection, including the following aspects.

Firstly, it has some reports to present for background information, then do some investigation about traditional pedestrian detection, people find that HOG SVM is more convenient to detect pedestrian on street. In addition, people find it is more faster to detect pedestrian on street. So it is quite popular among learners. Because YOLO is a one-time to predict multiple locations and categories of convolution neural network and it has good real-time performance, more easily for the pedestrian detection of video, so finally achieved the pedestrian detection technology based on YOLO v3, then test on existing data sets, by comparing the traditional pedestrian detection technology and the pedestrian detection technology based on the deep learning, validate the system for the real time and accuracy of the pedestrian detection.

Keywords: Pedestrian detection；Feature extraction；Deep learning；YOLO v3

目录

摘要 I

Abstract II

第一章 绪论 1

1.1 课题的背景和意义 1

1.2 行人检测的研究现状 1

1.3 本文的主要内容 2

1.4 本章小结 3

第二章 基于HOG SVM的行人检测算法 4

2.1 引言 4

2.2 传统的检测流程 4

2.3 HOG特征提取 6

2.4 SVM分类器 9

2.5 基于HOG特征提取 SVM分类器实现行人检测 12

2.6 本章小结 13

第三章 基于YOLO v3行人检测算法 14

3.1 引言 14

3.2 基于深度学习的行人检测简介 14

3.3 YOLO V3算法的检测流程 15

3.4 利用回归网络完成后续预测过程（YOLO v3） 15

3.4.1 Darknet介绍 19

3.4.2 YOLO V3检测模型的环境搭建 19

3.5 YOLO优缺点 22

3.6 本章小结 22

第四章 HOG SVM与YOLO算法实验对比 23

4.1 环境介绍 23

4.2 实验所用数据集 23

4.3 算法的主要参数 24

4.4 实验对比 27

4.5 本章小结 30

第五章 总结与展望 31

参考文献 32

致谢 34

请支付后下载全文，论文总字数：23552字