Abstract: when employing the machine vision system for monitoring the driver’s eyes fatigue state, due to image acquisition and image processing takes some time, reducing the system’s sampling rate, and the blink process is very short under normal circumstances, low sampling rate will loss the eyes of some key state information. This article directly studies the video image sequences, and uses the statistical methods to determine the maximum of open eye, then applies the PERCLOS method to determine the extent of eye fatigue. Study the varied sample intervals to impact on the PERCLOS value by sampling the video image sequences. Through six testers’ data show that, when the PERCLOS value reaches 15%, the sign of fatigue is obvious, with the deepening of fatigue, the PERCLOS value increases, adjusting the sample interval, it is found that when sample interval between the 40~120 ms, the PERCLOS value is relatively stable, the maximum relative error is 20.17%, but when the sample interval is greater than 120 ms, the PERCLOS fluctuation is high, the maximum relative error is 54.07%, which will affect the fatigue judgement.

Key words: PERCLOS, eyes open degree, determining fatigue

摘要： 在用机器视觉系统监测驾驶员的眼睛疲劳状态时，由于图像采集与图像处理消耗一定的时间，降低了系统的采样率，正常情况下，眼睛的眨动过程非常短暂，过低的采样率会漏掉眼睛某些关键状态的信息。因此直接对视频进行研究并提出用描述统计方法确定眼睛的最大睁开值，进而应用PERCLOS方法进行疲劳判定。对视频序列图像进行抽样，研究不同采样时间间隔对PERCLOS值的影响。通过六名测试者的数据表明：当PERCLOS值在15%附近时疲劳现象明显，随疲劳程度加深PERCLOS值升高，通过对视频序列图像抽样调整采样时间间隔，发现采样间隔在40 ms到120 ms之间时，PERCLOS值比较稳定，最大的相对误差为20.17%，采样间隔时间大于120 ms时PERCLOS值的波动幅度较大，最大相对误差达到54.07%，会影响疲劳判定结果。

关键词: PERCLOS, 眼睛睁开程度, 疲劳判定