基于卷积神经网络的人脸识别在疲劳驾驶检测中的应用

发布时间：2018-04-10 14:34

本文选题：深度学习 + 卷积神经网络　；参考：《广东技术师范学院》2017年硕士论文

【摘要】：卷积神经网络是一种源于人工神经网络的深度学习方法。它具有局部连接、权值共享的特点,同时能够实现特征的自动提取,它改善了传统模式识别方法中特征提取难的问题,因此卷积神经网络广泛应用于自然语言处理、语音识别、推荐系统、计算机视觉等领域。基于驾驶员外部特征的疲劳驾驶检测技术在多个方面取得了一定的进展,但是驾驶员脸部特征提取的方法有待进一步提高,同时驾驶员眼睛定位的时间较长,影响系统识别速率。论文将卷积神经网络应用于人脸识别,对瞳孔定位算法进行改进,有效地克服了原算法计算量大的问题,根据驾驶员眼睛在不同状态下宽与高比例不同的特点,实现了一种简单可行的眼睛状态判断方法,并通过PERCLOS算法对驾驶员的疲劳状态进行判定。应用卷积神经网络模型对ORL人脸库实验得到识别率为85%,识别平均时间为20ms。改进的Hough变换方法对驾驶员眼睛的定位准确率和平均时间分别为92%和29ms,对驾驶员眼睛状态的判断正确率为83.9%。论文使用的疲劳驾驶检测方法能比传统的检测方法取得更好的效果,设计基于人脸识别的疲劳驾驶检测原型系统,实现了驾驶员脸部特征检测、眼睛定位、眼睛状态判断、疲劳判定等功能。实验结果表明,系统对疲劳的识别率为87.5%,疲劳判断的响应时间为17ms,有较好的实际应用价值。
[Abstract]:Convolutional neural network is a kind of deep learning method derived from artificial neural network.It has the characteristics of local connection, weight sharing and automatic feature extraction. It improves the difficulty of feature extraction in traditional pattern recognition methods. Therefore, convolution neural network is widely used in natural language processing and speech recognition.Recommendation system, computer vision and other fields.Fatigue driving detection technology based on driver's external features has made some progress in many aspects, but the method of driver's facial feature extraction needs to be further improved, and the driver's eye location time is longer.It affects the recognition rate of the system.In this paper, the convolution neural network is applied to face recognition, and the pupillary location algorithm is improved, which effectively overcomes the problem that the original algorithm has a large amount of computation. According to the characteristics of the driver's eyes in different states, the width and the proportion of the eyes are different.A simple and feasible method for judging eye state is implemented, and the fatigue state of driver is judged by PERCLOS algorithm.By using convolutional neural network model, the recognition rate of ORL face database is 85% and the average recognition time is 20 Ms.The accuracy and average time of the improved Hough transform for eye localization are 92% and 29 msrespectively, and the correct rate for judging the eye state of the driver is 83.9%.The fatigue driving detection method used in this paper can achieve better results than the traditional detection method. A prototype system of fatigue driving detection based on face recognition is designed, which realizes driver's face feature detection, eye location, eye state judgment.Fatigue judgment and other functions.The experimental results show that the fatigue recognition rate of the system is 87.5 and the response time of fatigue judgment is 17mswhich has good practical application value.
【学位授予单位】：广东技术师范学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP183

【参考文献】