基于多特征融合的疲劳驾驶状态识别方法研究
来源:用户上传
作者:胡峰松 程哲坤 徐青云 彭清舟 全夏杰
摘要:针对交通安全中疲劳驾驶状态识别问题,使用单一的疲劳驾驶特征的方法识别率较低,本文提出一种基于面部多特征加权和的疲劳识别方法,通过人眼状态检测算法提取眼部疲劳参数,即持续闭眼时间、闭眼帧数比、眨眼频率,通过打哈欠状态检测得到打哈欠次数和打哈欠持续时间,通过头部运动状态分析得到点头频率,建立融合以上六个特征的驾驶疲劳状态检测模型来评估驾驶员的疲劳等级并进行相应的预警.实验测试数据选自NTHU驾驶员疲劳检测视频数据集的部分数据.经实验调整后,发现该方法的识别准确率较高,识别效果好.
关键词:驾驶安全;特征点定位;眨眼状态识别;多特征融合;疲劳识别
中图分类号:N39文献标志码:A
Research on Fatigue Driving State Recognition Method Based on Multi-feature Fusion
HU Fengsong,CHENG Zhekun,XU Qingyun,PENG Qingzhou,QUAN Xiajie
(College of Computer Science and Electronic Engineering,Hunan University,Changsha 41008 China)
Abstract:Aiming at the problem of fatigue driving state recognition in traffic safety,the recognition rate of using a single fatigue driving feature is low. This paper studies and proposes a fatigue recognition method based on the weighted sum of facial multi-features. The eye fatigue parameters,such as continuous eye closing time,eye closing frame ratio and blink frequency,are extracted by human eye state detection algorithm. The number and duration of yawning are obtained through yawning state detection,the nodding frequency is obtained through head motion state analysis,and a driving fatigue state detection model integrating the above six characteristics is established to evaluate the driver’s fatigue level and give the corresponding early warning. The experimental test data are selected from part of the NTHU driver fatigue detection video data set. After experimental adjustment,it is found that this method has high recognition accuracy and provide a good recognition effect.
Key words:driving safety;feature point positioning;blink of an eye state recognition;multiple feature fusion;fatigue recognition
S着汽车数量的迅猛增长,汽车在给人们生活带来快捷与便利的同时,频繁的道路交通事故也带来了惨重的经济损失,人民的生命安全受到了巨大的威胁,疲劳驾驶已成为全球一个严重而亟待解决的交通安全问题[1].研究表明,应用机器学习、模式识别等技术对疲劳驾驶状态检测识别及预警的方法效果较好[2-3],能有效预防道路交通事故的发生,疲劳驾驶检测技术已逐步从研究领域转移到工业应用领域并不断发展完善[4-5].
疲劳驾驶识别研究领域目前还存在很多难点,在不同的场景下如戴太阳镜、光照明喑等都会影响识别的准确率[4].疲劳驾驶状态信息包括眼部疲劳信息、嘴部疲劳信息和头部疲劳信息,识别方法可分为单一特征疲劳信息识别和多特征疲劳信息识别,而单一特征疲劳信息识别的准确率有待提高⑹本文主要根据多特征疲劳信息来进行疲劳驾驶状态识别.首先,基于SVM的睁闭眼状态识别算法判断眼部疲劳状态,然后通过嘴部高宽比和点头频率判断嘴部和头部疲劳状态,最后融合多特征进行疲劳驾驶状态识别.
1眼部疲劳状态判断
1.1人眼定位
使用由Kazemi和Sullivan提出的基于级联回归树的人脸关键点定位算法进行人眼定位。如图1 (a)中人脸特征点模型所示,根据图中关键点序号可以知道每个特征点的位置,如左眼的序号为36~4 右眼的序号为42~47.根据眼部特征点的序号,提取的左、右眼部区域如图1(b)所示的左、右长方形方框区域.其定位计算规则如下:
W=1.6×W,H=3×H(1)
式中:W为人眼特征点36到39之间的水平距离,H为人眼特征点37到41之间和38到40之间垂直距离的平均值,而W和H为定位的眼部区域的宽和高.
nlc202208291658
转载注明来源:https://www.xzbu.com/4/view-15438669.htm