现行的疲劳监测设备造价昂贵, 便携性差且有效性不高, 正面疲劳监测方案极大地影响了使用者的工作效率, 因此进行实用且准确的疲劳监测方案研究十分有意义。提出了新型的侧眼眼动疲劳监测眼镜并结合N-Range图像处理算法提高眼部定位的鲁棒性与疲劳监测的准确率。在侧眼图像处理过程中, 用N*N的卷积核及Sigmoid函数计算激活图, 采用OTSU阈值分割法对激活图进行阈值分割, 将小于阈值的像素点激活值置于0。据此计算水平方向与垂直方向的标准差投影, 从投影图中通过平均阈值法定位出人眼区域, 随后通过计算人眼高宽比表征眼部闭合度, 采用基于PERCLOS的P80疲劳评测方法为设备的使用者进行实时的疲劳分析。实验结果表明现行设备所存在的问题, 通过我们的方案得到了很好的解决, 即使在复杂的环境中, 本文的方法仍然表现出色, 疲劳判决准确度达到了94%。疲劳监测眼镜与N-Range算法可以实现在不影响使用者作业效率的前提下, 仍然保持着较高的疲劳监测精度, 具有较高的实用价值。

Current fatigue monitoring equipment is expensive, non-portable, and its algorithm has a low robustness in eye movement image processing. Existing front-end camera fatigue monitoring schemes have many limitations and greatly affect their users efficiency; as such, the development of portable and accurate fatigue monitoring systems is very challenging. In this paper, a new type of fatigue monitoring glasses and N-Range image processing algorithm is proposed to improve the robustness of eye location analysis and accuracy of fatigue detection. Real-time fatigue analysis is performed for users of the proposed mechanism according to the theory of percentage eye closure (PERCLOS) P80 fatigue assessment; it is considered that the P80 criterion is the most suitable for this study. In the process of side-eye image processing, an N-Range eye region extraction algorithm is proposed. The activation map is computed by an N*N convolution kernel, and it is segmented by the OTSU threshold segmentation method. The activation value of pixels smaller than the threshold is set at zero. Based on this, the standard deviation projection in the horizontal and vertical directions is calculated; the human eye region is located by the average threshold method from the projection map. The eye closure degree is measured by calculating the ratio of eye height to eye width and then counting the closure time. Experiments show that the aforementioned problems can be adequately solved by our method. Even in complex environments, the method in this paper still performs well, with fatigue judgment accuracy reaching 94%. The fatigue-monitoring scheme proposed in this paper, can be positively adapted under many uncertainties. As such, our fatigue monitoring glasses and N-Range algorithm can achieve high accuracy without affecting the efficiency of workers.