基于深度卷积神经网络的人眼检测

Eye detection based on deep convolutional neural networks

基于人眼检测的驾驶员疲劳检测研究中,受戴眼镜、光照强度和人脸姿态变化、闭眼以及面部遮挡等复杂背景影响,现有方法难以准确检测人眼位置的问题越来越严重,故提出一种基于深度卷积神经网络的人眼检测方法。对其进行网络优化及损失优化,把人眼检测作为一种回归问题来求解,实现了整个过程端到端,即从输入原始图片到最后的人眼类别和位置的输出。该方法在ORL人脸数据库中全部图像的人眼检测准确率为98.39%,在AR人脸数据库中没有戴墨镜的人眼检测准确率为95.15%,实验结果验证了所提方法的有效性、高准确率和较强的泛化能力。

In the driver fatigue detection research based on eye detection,the problem that it is difficult for the existing methods to accurately detect the eye position is becoming more and more serious due to the influences of complex backgrounds such as glasses wearing,light intensity,facial pose variation,eye closure,and facial occlusion.Therefore,an eye detection method based on deep convolutional neural networks is proposed.Network optimization and loss optimization are performed for the method.The eye detection is solved as a regression problem,so as to realize a whole end-to-end process which is from input of original images to output of eye classification and position at last.The method has an accuracy rate of 98.39%for eye detec-tion of all images in the ORL face database,and an accuracy rate of 95.15%for detection of eyes(without wearing sunglasses)in the AR face database.The experimental results verified the effectiveness,high accuracy and strong generalization capability of the proposed method.