提出基于卷积神经网络的驾驶行为分析算法,该算法在人脸定位的基础上实现了驾驶员的疲劳检测和行为检测。针对疲劳检测任务,探究了卷积神经网络的不同感受野对疲劳检测精度的影响,并得到了疲劳检测模型的最佳结构。针对行为检测任务,考虑到不同行为对应作用域的大小不同,提出了一种基于多尺度特征的多支路注意力网络模型,该模型通过提取多尺度特征实现了多尺度分类,并且使用注意力机制来强化判别特征。实验结果证明,该方法能够与多种主流卷积神经网络模型相结合并有效提升驾驶行为分析的准确率。

This paper proposes a driving behavior analysis algorithm based on convolutional neural network, which realizes the driver's fatigue detection and behavior detection based on face location. Aiming at fatigue detection task, this paper explores the influences of different receptive fields of convolutional neural network on the accuracy of fatigue detection and obtains the optimal structure of the fatigue detection model. For the behavior detection task, considering that the corresponding scope sizes of different behaviors are different, a multi-branch attention network model based on multi-scale features is proposed, which realizes multi-scale classification by extracting multi-scale features and exploits attention mechanism to strengthen distinguishing features. Experimental results show that this method can be combined with a variety of mainstream convolutional neural network models and effectively improves the accuracy of driving behavior analysis.