Abstract: The main method of fatigue detecting can be divided into two part, one would be detecting the behavior of body which includes the head and stuff. The other would be detecting the biological signals including the state of mouth and eyes or the brain signals. The common and the disadvantages of these processes is the judgement that only can be made after getting into the fatigue state. Serious accident might happen if the driver starts braking delay tenths of second. Aiming at this disadvantage, this paper presents a method about forecasting the next period fatigue state according to the current data. It can prevent the accident happening effectively. It locates the eyes in multi-angle by using the horizontal gray-level projection and the gray value standard deviation, counts the eye-state-threshold by using a simple and accurate method presented by this paper. Markov chain would be used to forecast the fatigue state. The result shows that this method has high accuracy in forecasting with a good real-time.

Key words: eye location, multi-angle, Sobel edge extraction, Markov chain, forecast

摘要： 对人类疲劳的检测，目前分为检测身体活动，包括头部等一些身体行为和检测生物信号，包括眼睛嘴巴状态或脑电波信号等，这些方法的共同点和弊端是在当事人进入疲劳状态后才能作出判断，但对于驾驶员制动延迟零点几秒也可能酿成重大事故的情况，以往的方法显然是不适应的，针对此弊端，特别提出一种根据当前数据预测下一个时间段的精神状态的方法，能有效地防止事故的发生。利用灰度投影法与灰度变化标准差的结合完成多角度眼睛的定位，根据提出的一种简便且准确的样点提取法来计算眼睛状态对应的阈值，利用马尔科夫链算法对司机的精神状态进行判断和预测。实验结果表明，该方法预测准确率高并有很好的实时性。

关键词: 人眼定位, 多角度, Sobel边缘提取, 马尔科夫链, 预测