Gentle-Adaboost在红外视频驾驶员疲劳检测中的应用研究  

Research on Fatigue Detection by Infrared Videos Based on Gentle-Adaboost

作　　者：朱喜燕[1] 许勇[1] 汝正阳[1] 

机构地区：[1]桂林电子科技大学电子工程与自动化学院,广西桂林541004

出　　处：《计算机测量与控制》2015年第4期1414-1417,共4页Computer Measurement &Control

摘　　要：眼睛状态的检测是驾驶员疲劳检测的关键,但夜间复杂的行车环境导致眼睛状态不易检测,针对这种情况,首先用GentleAdaboost算法对大量红外样本训练得到人脸和眼睛分类器,用分类器对驾驶员进行面部和眼睛检测,并提出了采用高斯模型对眼睛区域的垂直积分投影分析得到眼睛睁闭状态的方法,进而计算PERCLOS和EBN,对驾驶员的精神状态进行检测;通过定位人脸以缩小眼睛的搜索区域,不仅可以提高眼睛的检测率,还可提高检测速度;在Visual Studio 2012和Opencv 2.4.4中对该系统进行仿真,验证了其有效性和实时性。Eyes closure evaluation is very important for driver fatigue detection, however, the complex driving environment in night makes the work hard. Therefore, Gentle Adahoost is adopted in this paper to train samples of NIR images for face and eye classifiers, which are used to detect the face and eyes. Horizontal and vertical integral projections are proposed to locate the eyes more precisely, then the eye shape is evaluated by analyzing the vertical integral projections with Gaussian model, which can estimate the fatigue state of driver. The method has been proved effective with proper real--time performance on Visual Studio 2012 and Opencv 2.4.4.

关 键 词：Gentle-Adaboost 积分投影 高斯模型 PERCLOS EBN 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]