Eye blink detection using CNN to detect drowsiness level in drivers for road safety
Eye blink detection using CNN to detect drowsiness level in drivers for road safety
复制标题
DOI：
10.11591/ijeecs.v22.i1.pp222-231
复制
发表时间：
2021-04
影响因子：
--
通讯作者：
Pothuraju Vishesh;S. Raghavendra;SantoshKumar Jankatti;Rekha
中科院分区：
文献类型：
--
作者：
Pothuraju Vishesh;S. Raghavendra;SantoshKumar Jankatti;Rekha
眨眼是一种正常的身体功能，它是眼睑的半自动快速闭合。通过眼睑的动态折叠来检查特定眨眼。这是眼睛的一个重要功能，有助于眼泪的传播，并消除角膜浅层的刺激物。在这项研究工作中，我们利用卷积神经网络，深度学习概念和图像处理来检测驾驶员的困倦程度。为了训练眨眼检测模型，使用mobilenet V2作为基础。用于训练的损失函数是RMSprop，优化器是二进制交叉熵。利用dlib人脸标志点对检测到的人脸进行感知和预处理。用于训练模型的数据集选自南京航空航天大学的“晓阳滩”。实验结果表明，该方法的准确率达到97%。开发的原型作为进一步开发这一过程的基础，以实现更好的道路安全。
Blinking is a regular bodily function and it is the semiautomatic fast closing of the eyelid. A specific blink is examined by dynamic folding of the eyelid. It is a vital function of the eye which helps in spread of tears across and eliminates irritants from the shallow of cornea. In this research work we made use of convolution neural network, the deep learning concepts and image processing to detect drowsiness level in drivers. To train the blink detection model the mobilenet V2 is used as base. The loss function used for training was RMSprop and the optimizer is binary cross entropy. The dlib facial landmark was exploited to perceive and pre-process the detected faces. The dataset used for the training model is selected from the “Xiaoyang Tan” of nanjing university of aeronautics and astronautics. Based on the experimental outcome the projected method achieves an accuracy of 97%. The prototype developed serves as a base for further development of this process to achieve better road safety.