基于眼睛状态识别的驾驶员疲劳实时监测

Real-time Monitoring of Driver Fatigue Based on Eye State Recognition

提出了一种基于眼睛状态识别的驾驶员疲劳状态实时监测的方法。为了实现眼睛状态的检测,首先通过计算累计背景和当前帧的差分图像的质心确定脸部范围,然后通过二值化和轮廓检测确定眼睛的搜索区域。在利用启发式规则进行筛选定位之后,计算眼睛骨架曲线和两眼角连线之间的距离得到眼睛的睁开程度。通过计算相应的疲劳指标如PERCLOS、平均睁眼程度、最长眼睛闭合时间来推测驾驶员的疲劳状态。以驾驶员面部视频的主观评分作为评价依据对检测方法进行评价,结果显示上述3个指标在不同的疲劳等级下均存在显著性差异,通过对不同指标的融合可达到较好的疲劳检测准确率。

A scheme of real-time monitoring of driver fatigue based on eye state recognition is presented. First, the difference images obtained from the background and current images are used to identify the center of face region. Then binarization and contour detection are used to identify the region of interest of the eye. After using heuristic rules to screen out the contour, the profile of eye rim can be extracted, based on which the opening level of eye is obtained as one of the three parameters chosen to be the indicators of driver fatigue. The three indicators are percent eyelid closure （ PERCLOS）, average opening level, and maximum closure duration of eye. They are e- valuated with subjective rating by groups of laboratory technicians on the video of drivers＇face images. The results show that all the three above-mentioned indicators can well represent driver fatigue, and the fusion of three indicators can get even higher accuracy rate in driver fatigue detection.