论文部分内容阅读
基于计算机视觉的疲劳检测具有低侵入性、低成本的优点,然而光照变化、面部表情、复杂背景等仍然对检测率造成很大的影响。以卷积神经网络为代表的深度学习以其较强的特征提取能力和鲁棒性在模式识别领域取得了成功的应用。本文提出了一种基于级联卷积神经网络(CNN)结构的疲劳检测算法。首先训练第1级网络实现人眼与非人眼的分类,使网络充分学习人眼特征,当输入目标图像时,人眼区域能快速从第一级网络特征图中分离出来;然后将人眼图像传送给第2级网络检测眼部特征点位置,计算眼睛张开度并以此判断测试者眼睛状态,构造疲劳检测模型;最后根据连续多帧的眼睛状态序列,判断测试者是否处于疲劳状态。在检测误差为5%时,眼部4个特征点的平均检测正确率为93.10%,单点检测正确率最高可达97.14%。测试结果表明,在本文提出方法下眼睛的清醒和疲劳状态有明显的不同,证明本文提出的方法有效可行,具有较好的应用前景。
Fatigue testing based on computer vision has the advantages of low invasiveness and low cost. However, light changes, facial expressions, complex backgrounds and the like still have a great impact on the detection rate. Deep learning represented by convolutional neural networks has been successfully applied in pattern recognition because of its strong feature extraction ability and robustness. This paper presents a fatigue detection algorithm based on cascaded convolutional neural network (CNN) structure. Firstly, the first level network is trained to classify the human eye and the non-human eye so that the network fully learns the human eye features. When the target image is input, the human eye area can be rapidly separated from the first-level network feature map; and then the human eye The image is sent to the second level network to detect the location of the eye feature points, calculate the eye opening degree and judge the state of the eye of the tester to construct the fatigue detection model. Finally, judge whether the tester is in a state of fatigue according to the continuous multi-frame eye state sequence. When the detection error is 5%, the average detection accuracy of the four characteristic points in the eye is 93.10%, and the accuracy of single point detection is up to 97.14%. The test results show that there is a clear difference in the awakeness and fatigue state of the eyes under the proposed method, which proves that the proposed method is effective and feasible and has good application prospects.