基于人眼状态信息的疲劳检测方法研究

Research on Fatigue Detection Method Based on Human Eye State Information

为实现复杂环境下的疲劳驾驶检测与预警,论文利用红外穿透来解决疲劳驾驶中复杂光线和墨镜遮挡等的影响,论文使用改进的Yolov3-tiny算法来进行驾驶员的疲劳检测,先对聚类算法进行改进,然后增加检测尺度进行多尺度融合,来降低小目标的误检和漏检,并通过深度特征融合来进行跨层连接。然后进行人眼特征点检测,再根据PERCLOS准则来判断是否疲劳,实验结果表明该算法能快速在嵌入式设备上进行疲劳驾驶检测。

For fatigue detection and early warning in a complicated environment,this paper uses infrared penetration to solve the influence of complex light and shades in fatigued driving,this article uses the improved Yolov3-tiny algorithm for driver fatigue detection,first the clustering algorithm is improved,and then the multi-scale fusion detection measure is increased,the error checking of the small target and residual is reduced,and through the depth of feature fusion,cross layer connection is performed.Then the key points of the eyes are detected and the aspect ratio of the eyes is calculated. Then the fatigue is judged according to PERCLOS criterion. The experimental results show that the algorithm can quickly detect the fatigue driving on embedded devices.