基于面部特征的疲劳驾驶检测方法研究

Research on Fatigue Driving Detection Method Based on Facial Features

疲劳驾驶是遥成交通事枚的主要原因之一,通过检测驾驶员是否覆•势来预防交通事故并有效保障行车姿全,具有非常重要的意义。随着近几年深度学习技术的兴起,基于驾驶员面部特征的疲•势检测方法又可以划分为传统的疲•势检测方法和基于深度学习的疲劳榜测方出。本丈研堯青先自建卖验数据集;接着使用相同的数据集分别训练原始的DeepID网络棋型以及使用了DropoutConv算冻的DeepID<型;之后分别测试两个模型分别在训练数据集和独立疲•势教据集上的正确率,使用模型在两个数据集上正确率的盖距来评价网络模型的过拟合程度;最后证明了DropoutConv算法对于減轻网络模型过拟合的有效性,该算法可以降低模型对于训练软据集的过度拟合,并提高模型的泛化性能.

Fatigue driving is one of the main causes of trafficaccidents. It is of great significance to detect whetherthe driver is tired to prevent traffic accidents andeffectively ensure driving safety. With the rise of deeplearning technology in recent years, fatigue detectionmethods based on drivers, facial features can be dividedinto traditional fatigue detection methods and deeplearning-based fatigue detection methods. In this paper,the self-built experimental data set is firstly presented.Then the original DeepID network model is trained bythe same data set and the DeepID model is trained byDropoutConv algorithm. Then, the accuracy of the twomodels on the training data set and the independentfatigue data set was tested respectively. At last, it isproved that the algorithm of DropoutConv is effective inreducing the overfitting of network model, which canreduce the overfitting of model to training data set andimprove the generalization performance of model.