论文总字数：20930字

摘 要

路怒症已成为影响安全驾驶的一个重要因素，它是由于生活压力心理原因等多方面原因，以看到别人违章为直接导火索，引起驾驶员愤怒的情绪，从而引发交通事故、危害人身安全。近年来越来越多人开始从事驾驶员路怒症相关的研究。人的面部表情是表现人类情绪的重要途径，通过开发监控系统，实时检测并显示驾驶员表情，并对驾驶愤怒情绪进行识别，及时的进行提醒，对防止驾驶员路怒症带来的交通事故具有重要意义。

本课题的主要工作是在提取脸部特征的基础上，设计并训练深度学习网络模型，实现人脸表情特征提取，选择适合的分类器，对路怒症相关表情进行识别。在驾驶员人脸检测方面，本文采用基于像素相似性度量的检测方法，提取人脸haar−like特征，利用Adaboost算法将haar−like特征进行组合达到强学习的目的，同时级联多个强分类器来提高准确率。在面部表情识别方面，采用基于深度学习的方法，利用DeepNN网络模型，结合softmax多分类器对驾驶员愤怒情绪有关的愤怒、厌恶表情进行识别。

本文对公开数据集fer2013、MMI和KMU-FED进行了仿真测试，并与多种先进的方法进行了性能比较。平均识别率Fer2013为52.1%，MMI为66%，KMU-FED为76.7%，其中愤怒和厌恶可归结于路怒症，在检测到此类表情时，对司机进行提醒。

关键词：表情识别，人脸检测，特征提取，路怒症

Abstract

Road rage has become a significant factor influencing safe driving. It is caused by the mental reasons of life pressure and other factors. Seeing others violate the regulations is a direct trigger to cause drivers' anger, thus causing traffic accidents and endangering personal safety. In recent years, more and more person start to work on the research related to driver road rage Human facial expression is a significant way to express human emotions. It is of great significance to exploit a supervisory system to detect and display the driver's facial expression in real time, identify the driver's anger emotion, and timely remind and relieve it in order to prevent traffic accidents caused by driver's road rage.

The main work of this subject is to design and train the network model based on the extraction of facial features, so as to achieve the extraction of facial expression features, select the appropriate classifier and classify and recognize emotions.In the driver face detection part, this paper uses the discover method based on pixel similarity measurement of OpenCV to extract the haar-like features of the face. Adaboost algorithm is used to combine multiple haar-like features to achieve a better strong learning algorithm, and multiple strong classifiers are cascaded to increase the recognition accuracy. In the facial expression recognition part, softmax multi-classifier based on DeepNN network model was used for driver facial expression recognition.

The fer2013 database, MMI and kmu-fed Facial Expression of Drivers (kmu-fed) database of Keimyung university were evaluated experimentally in this paper, and their performance was compared with the latest methods.Fer2013 was 52.1%, MMI was 66%, and kmu-fed was 76.7%, which obtained good facial expression recognition performance.

There were common problems with dnn-based facial expression recognition in the experiment: it could not operate on low-specification equipment and required a large amount of data to operate effectively.In theory, fer2013 should be able to get 65 percent accuracy, but the experiment was much lower.

Key Words：emotion recognition, face detection, characteristic extraction, road rage

目录

Abstract 3

第一章 绪论 7

1.1 概述 7

1.2 研究现状 7

1.3 本文的主要工作 9

第二章 人脸检测和情感识别的相关方法及理论 10

2.1面部表情识别方法概述 10

2.1.1传统面部表情特征的提取方法 10

2.1.2基于深度学习的面部表情特征的提取方法 11

2.2卷积神经网络 12

2.3常见分类器 14

第三章 基于haar-like特征和𝐀𝐝𝐚𝐛𝐨𝐨𝐬𝐭分类器的人脸检测 15

3.1引言 15

3.2基于haar-like特征和Adaboost分类器的人脸检测原理 15

3.3人脸检测算法整体设计 19

3.4实验结果与分析 21

3.4.1实验环境搭建 21

3.4.2结果分析 22

第四章 基于DeepNN网络的驾驶员路怒表情识别 24

4.1基于DeepNN网络的驾驶员路怒表情识别原理 24

4.2基于DeepNN网络的驾驶员路怒表情识别整体设计 26

4.3实验结果与分析 27

4.3.1实验环境搭建 27

4.3.2数据集介绍及预处理 28

4.3.3参数设置 30

4.3.4性能分析 30

第五章 总结与展望 33

5.1总结 33

5.2系统评价 33

5.3未来展望 33

参考文献 34

致谢 37

附录主程序代码 38

第一章 绪论

1.1 概述

驾驶员路怒症又称阵发性暴怒症，它是由于生活压力心理原因等多方面原因，以看到别人违章为直接导火索，引起驾驶员愤怒的情绪，从而引发交通事故、危害人身安全。路怒症以成为影响安全驾驶的一个重要因素，近年来越来越多人开始从事驾驶员路怒症相关的研究。人的面部表情是表现人类情绪的重要途径，通过开发监控系统，实时检测并显示驾驶员表情，并对驾驶愤怒情绪进行识别，及时的进行提醒和缓解，对防止驾驶员路怒症带来的交通事故具有重要意义。

驾驶员路怒情绪识别属于情感计算范畴^[1],心理学家梅赫拉比安的研究表明，面部表情包含高达55%的情感信息。利用面部表情，我们将可以得到体现驾驶员心理状态的信息。如果能够做到驾驶员实时人脸检测，并判断驾驶员是否有愤怒情绪，将对驾驶员路怒识别有莫大的帮助。

请支付后下载全文，论文总字数：20930字