面部多特征融合的列车司机疲劳检测方法研究

Fatigue Detection Method for Train Drivers Based on Facial Multiple Feature Fusion

为保障列车安全运营,提出一种基于面部多特征融合的列车司机疲劳检测方法。采用HOG特征实现人脸识别,ERT算法实现特征点定位。针对不同司机面部特征差异导致固定阈值无法适用于每位司机的问题,提出基于K-means++法的人眼自适应阈值算法。考虑图像前后帧的相关性,利用相邻两帧图像瞳孔与内眼角差值计算眼动速率作为判断指标。针对司机可能出现短暂睁眼而非真正清醒的情况,通过模糊推理系统将眼睛开合度、嘴巴开合度和眼动速率三个指标融合以提高疲劳检测的准确率,同时实现疲劳分级。实验表明,该方法准确率达95%。

In order to ensure the safe operation of trains,a method for fatigue detection of train drivers based on facial multi-feature fusion was proposed.First,HOG features were used to realize face recognition while ERT algorithm was adopted to locate facial feature points.Aiming at the problem that different drivers’facial features make a fixed threshold unable to apply to each driver,an adaptive threshold algorithm for eyes based on K-means++was proposed.Considering the correlation between the previous and following image frames,the difference between pupils and inner eye corners in two adjacent frames was used to calculate the eye movement rate as a judgment factor.In response to the situation where drivers may briefly open his eyes during the drowsiness when they are actually not alert,the three indicators of eye opening and closing degrees,mouth opening and closing degrees,and eye movement rate were merged to improve the fatigue detection accuracy through fuzzy inference system,and to realize fatigue classification at the same time.The experiments show that the fatigue detection method in this paper has a high accuracy rate of 95%.