摘要
驾驶疲劳监测对算法实时性及可靠性要求高,提出一种快速人眼定位及状态分析算法。首先采用Adaboost分类器在视频序列中定位人脸,并给出图像自商融合方法增强算法对光照的适应能力;然后利用灰度投影划分人眼区域,并通过Bernsen局部自适应二值化方法提取眼睛候选区;其次对候选区进行图像矩描述,通过椭圆参数分析及上下文信息实现眼睛定位;最后针对疲劳判定指标特点,采用椭圆短轴参数表征眼睛睁开高度实现眼睛状态分析。实验表明,算法对光照鲁棒性强,平均正确定位率达到98.34%,监测速度达到28 f/s(帧/秒),算法适用于疲劳判断指标,可应用于疲劳监测系统,具有较高的实用价值。
System of drive fatigue detection has high requirements to the real-time and reliability of algorithm, an algorithm of quickly eye location and state recognition is proposed in this paper. Firstly, face is located by Adaboost classifier in video sequence, and a method of image self quotient is given to enhance adaptability to illumination. Secondly, eyes region is divided out by gray projection and eyes candidate region is extracted by Bemsen ＇ s binary method. Thirdly, image second moment is used to characterize the candidate region and realize the eyes position by analyzing ellipse parameters and context information. Finally, aiming to the features of fatigue detection judgment standards, short axis parameters of ellipse is used to describe the closure of eye. Experiments showed that, the algorithm has strong robustness to illumination, the average correct position rate is 98.34% , and the detection rate is 28 f/s. The algorithm can be used in driver fatigue detection system, and the application value of the algorithm is high.
出处
《电视技术》
北大核心
2014年第7期194-198,共5页
Video Engineering
基金
国家科技重大专项课题资助项目(2009ZX02308-004)
关键词
PERCLOS指标
自商融合
图像矩
上下文信息
PERCLOS standard
image fusion of self quotient
image moment
context information
