机车司机驾驶状态监测方案的研究  

Research on monitoring plan of fatigue driving status for locomotive driver

作　　者：王莹[1] 赵守国[1] 王静[2] 王珂[2] 

机构地区：[1]北京交通大学计算机与信息技术学院,北京100044 [2]中国铁道科学研究院电子计算技术研究所,北京100081

出　　处：《铁路计算机应用》2012年第12期7-10,共4页Railway Computer Application

摘　　要：通过判断手势的规范性与标准性,对机车司机的驾驶状态进行监测。首先对机车司机驾驶过程的视频文件进行一定的预处理,通过混合高斯模型、帧间差分法及背景减除法,得到不包含背景噪声的规定手势图像;然后,通过肤色模型的转换对手势的图像进行准确提取;再通过腐蚀与扩张等方法对其进行修正,从而得到清晰准确的二值图像;最后采用形状上下文算法、HOG特征和AdaBoost分类器等对手势的二值图像进行目标的识别与综合的判定,进而判断出机车司机的驾驶状态,对其进行监测。实验数据表明,视频文件经上述的运算预处理,可以较为准确地对机车驾驶的视频文件进行驾驶状态的监测。Through determining whether the gestures were normative or standard, the locomotive driver＇s driving status could be monitored. First of all, the video files should be preprocessed. By Gaussian mixture model, frames subtraction method and background subtraction method, the gestures frame without background was got then, by skin-color model conversions, the gesture was more accurate extracted from the frame. The method of corrosion and expansion were used to make amendments and get the gesture clearer. Last shape context, HOG feature and AdaBoost classifiers were needed to identify and judge comprehensively. Through judging the standard gestures, the driver＇s driving status was monitored. Experimental data showed that, through a series of algorithms of calculation, the locomotive driver＇s driving states could be determined more accurately.

关 键 词：手势识别 肤色模型 HOG特征 AdaBoost多分类器 形状上下文 

分 类 号：U268.48[机械工程—车辆工程] TP39[交通运输工程—载运工具运用工程]