基于面部行为分析的驾驶员疲劳检测方法研究

发布时间：2018-02-25 21:00

本文关键词： 人脸检测状态识别 CNN PERCLOS 疲劳检测　出处：《天津工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来交通事故频发,给国家和个人带来了严重的财产损失。研究表明,疲劳驾驶是目前引发交通事故的主要原因之一,已经引起许多国家和政府的重视,因此准确快速的驾驶员疲劳检测的研究具有重要的意义。基于机器视觉的检测方法以其非接触性、实时性等优点,成为驾驶员疲劳检测的一个重要方法。眼睛和嘴部等状态的检测是疲劳检测方法中的重要步骤,但是墨镜遮挡及光照变化会对其产生影响。针对以上问题,本文使用红外采集设备对驾驶员面部图像进行采集,提出一种基于面部行为分析的驾驶员疲劳检测方法,其中主要研究内容包含人脸检测及跟踪、眼睛和嘴部区域检测、面部状态识别及疲劳检测等。首先,通过基于AdaBoost的检测检测算法进行驾驶员面部检测,为了提高检测速度及准确率,本文结合基于KCF(Kernelized Correlation Filter)的跟踪算法,对检测到的人脸区域进行快速跟踪;其次,通过级联回归的方法定位面部关键点,根据关键点位置提取眼睛和嘴部区域;最后,采用CNN(Convolution Neural Network)网络模型对提取出的眼睛和嘴部区域进行状态识别,得到眼睛和嘴部状态后,计算 PERCLOS(Percentage of Eyelid Closure Over the Pupil Over Time)、眨眼频率及打哈欠参数等,通过结合多个疲劳参数对驾驶员的疲劳状态进行检测。实验结果表明,该方法在佩戴墨镜情况下能够更准确的检测眼睛和嘴部状态,进而得到更准确的疲劳参数。与仅采用PERCLOS参数的方法相比,通过结合多个疲劳参数能够得到更为准确的结果。
[Abstract]:The frequent occurrence of traffic accidents in recent years has brought serious property losses to countries and individuals. Studies show that fatigue driving is one of the main causes of traffic accidents at present and has attracted the attention of many countries and governments. Therefore, the research of accurate and fast driver fatigue detection is of great significance. The detection method based on machine vision has the advantages of non-contact, real-time and so on. The detection of eye and mouth is an important step in the fatigue detection method, but sunglasses occlusion and light change will have an impact on it. In this paper, the driver's facial images are collected with infrared acquisition equipment, and a driver fatigue detection method based on facial behavior analysis is proposed. The main research contents include face detection and tracking, eye and mouth region detection. First of all, the driver face detection is carried out through the detection algorithm based on AdaBoost. In order to improve the detection speed and accuracy, this paper combines the tracking algorithm based on KCF(Kernelized Correlation filter. The detected face region is tracked quickly. Secondly, the key points of the face are located by cascading regression, and the eye and mouth regions are extracted according to the key points. Finally, The CNN(Convolution Neural network model was used to recognize the state of the extracted eye and mouth regions, and then the PERCLOS(Percentage of Eyelid Closure Over the Pupil Over time, blink frequency and yawning parameters were calculated. The test results show that the method can detect the state of eyes and mouth more accurately in the case of wearing sunglasses. More accurate fatigue parameters can be obtained, and more accurate results can be obtained by combining multiple fatigue parameters compared with the method using only PERCLOS parameters.
【学位授予单位】：天津工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;TP391.41

【参考文献】