采用ELM的基于眼部状态的驾驶员疲劳检测方法的研究

摘要

目前，交通安全问题成为世界性的广泛话题，根据统计，大约有40％的交通安全问题来自疲劳驾驶。研究驾驶员疲劳监测方法，用以预防交通事故的发生，具有非常重要的现实意义。
　　本文在对现有的驾驶员疲劳检测相关技术进行深入分析的基础上，对视频分析方法进行了深入的研究，提出了一种采用ELM(Extreme LearningMechine，极限学习机)技术的、基于眼部状态的驾驶员疲劳监测方法。首先，通过基于皮肤颜色的视图分割方法来进行检测驾驶员的脸部。接着采用KLT(Kanade-Lucas-Tomassi)算法，对脸部检测数据进行跟踪并截取出每帧图像。然后，通过使用一个积分映射函数用来估测眼睛的区域，以检测眼睛张开或者闭合的状态。进而计算一组帧图像中的眼睛开闭程度距离，并基于上述数据进行特征抽取（眨眼频率和眼睛关闭时间）。由于在面向视频的计算中进行多次迭代计算，数据量将非常庞大而且时间复杂度很高。为了降低时间复杂度，减少计算量，引入了滑动窗口的数据处理方式进行实时计算。最后，本文采用提取出的特征数据对ELM进行训练，从而对驾驶员的疲劳状况进行检测和分类。
　　本文利用实际的网络摄像头的实时视频数据进行了实验研究，结果表明，本文所提出的方法在保证实时性要求的前提下，能够有效地提取驾驶员的眼部特征，准确地判断驾驶员的工作状态。

目录

声明

ABSTRACT

摘要

CONTENT

LIST OF FIGURES

Chapter 1： INTRODUCTION

1．1 Problem statement and thesis background

1．2 Thesis objectives

1．3 China and Abroad research status

1．3．1 China researches status

1．3．2 Abroad researches status

1．4 Related Fatigue Detection Techniques

1．4．1 Detection based on physiological characteristics

1．4．2 Detection based on vehicle behavior characteristics

1．4．3 Detection based on driver behavior characteristics

1．5 The thesis outlines

1．6 Chapter summary

Chapter 2：RELATED IMAGE PROCESSING TOOLS

2．1 Image representation

2．2 brayscale transforms

2．3 Histogram and Histogram Equalization

2．3．1 Histogram

2．3．2 Histogram equalization

2．4 Image Binarization

2．5 Image Filtering

2．5．1 Gaussian kernel filter

2．5．2 Median filter

2．6 Morphological image processing

2．6．1 Erosion and Dilation

2．6．2 Opening and Closing

2．7 Chapter summary

Chapter 3 IMAGE ACQUISITION AND FACE DETECTION TECTHNOLOGIES

3．1 Image acquisition

3．2 Face detection techniques

3．2．1 Template matching method

3．2．2 Feature invariant method

3．2．3 Appearance-based method

3．2．4 Knowledge-based methods

3．2．5 Skin color-based segmentation

3．3 Proposed face detection and tracking method

3．3．1 Skin color based segmentation for face detection

3．3．2 Kanade-Lucas-Tomasi(KLT)algorithm for face tracking

3．3．3 Result of the proposed method

3．4 Chapter summary

Chapter 4：EYE LOCALIZATION AND DETECTION TECTHNOLOGIES

4．1 Eye detection technologies

4．2 Edge detection methods

4．2．1 Sobel edge operator

4．2．2 Robert’s edge operator

4．2．3 Prewitt edge operator

4．2．4 Laplaeian of Gaussian operator

4．2．5 Canny edge operator

4．3 Eyes detection

4．3．1 Eyes area localization

4．3．2 Eyes detection by integral projection

4．4 chapter summary

Chapter 5：Fatigue detection and extreme learning machine

5．1 Features extraction(eye closure time，blinking frequency)

5．2 Extreme learning machine Principe

5．3 Experimental analysis

5．4 Hardware software environment

5．4．1 MATLAB software

5．4．2 Hardware

Chapter 6 CONCLUSION AND FUTURE WORK

6．1 Conclusion

6．2 Future work

REFERENCES

ACKNoWLEDGMENTS