基于卷积神经网络的疲劳检测改进算法OA北大核心CSTPCD
AN IMPROVED ALGORITHM FOR FATIGUE DETECTION BASED ON CNN
为了解决当前的疲劳检测算法准确率低或实时性差的缺点,提出一种改进的卷积神经网络疲劳检测算法.使用HOG检测算法结合KCF跟踪算法对采集的人脸进行检测和跟踪;随后调用Dlib库进行脸部关键点的提取;通过引入可变形卷积神经网络对提取的眼部和嘴部进行状态识别;通过CEW和YAWDD数据集进行测试,疲劳检测准确率达到94.36%.实验表明,与当前的疲劳检测算法相比,提出的方法能够实时地检测驾驶员疲劳,并且具有较高的准确率.
In order to solve the shortcomings of low accuracy or poor real-time performance of current fatigue detection algorithms,an improved convolution neural network fatigue detection algorithm is proposed.HOG detection algorithm combined with KCF tracking algorithm was used to detect and track the collected faces.The Dlib library was called to extract the key points of the face.A deformable convolution neural network was introduced to identify the extracted eye and mouth states.This algorithm was tested by CEW and YAWDD data set.The accuracy of fatigue detection reaches 94.36%.Experiments show that compared with the current fatigue detection algorithms,the proposed method can detect driver fatigue in real time with high accuracy.
周先春;邹清宇;陆滇
南京信息工程大学电子与信息工程学院 江苏南京 210044||江苏省大气环境与装备技术协同创新中心 江苏南京 210044南京信息工程大学电子与信息工程学院 江苏南京 210044
计算机与自动化
人脸检测Dlib可变形卷积状态识别疲劳检测
Face detectionDlibDeformable convolutionState recognitionFatigue detection
《计算机应用与软件》 2024 (006)
156-160,168 / 6
国家自然科学基金项目(11202106,61302188);江苏省大学生创新创业训练计划项目(202010300128P).
评论