【摘要】 现有的驾驶员行为检测、疲劳检测模型存在误差较大、检测效果单一等问题，是由于多变的光线以及数据集多为图片所导致。针对以上问题，提出一种结合时空信息和自注意力机制的图卷积网络模型。模型采用注意力池化图卷积网络(Attention Pooling Graph Convolution Net, APGCN),在时空图卷积网络中的全局池化层中引入自注意力机制，并在最后一层卷积加入有效通道注意力模块(Efficient Channel Attention, ECA)。使得模型可以在同时考虑节点特征和图的拓扑结构的条件下实现分层池化，并且在不改变特征图尺寸的情况下，对输入的特征图进行了通道特征加强。实验结果表明，模型准确率提高了3.16%,检测速度提高了3.5帧/s,与其它相关方法的对比也证明了其有效性。更多还原

【Abstract】 The existing driver behavior detection and fatigue detection models have problems such as large errors and single detection effects, which are caused by changing light, most of the data sets are pictures. To solve this problem, a graph convolutional network model combining spatio-temporal information and a self-attention mechanism is proposed. The model uses an Attention Pooling Graph Convolution Net(APGCN)and introduces a self-attention mechanism into the global pooling layer of a spatio-temporal graph convolution network. An Efficient Channel Attention(ECA)module is added to the last convolution layer. The model can realize hierarchical pooling under the condition of considering the node characteristics and the topology structure of the graph at the same time, and the channel characteristics of the input feature map are strengthened without changing the size of the feature map. The experimental results show that the accuracy of the model is improved by 3.16%,and the detection speed is improved by 3.5 frames per second. The comparison with other related methods also proves the effectiveness of the proposed method.更多还原