双线性融合网络的驾驶员分心行为识别

准确识别驾驶员的分心行为能够从源头上减少交通事故的发生。传统的识别方法类别少,准确率不高,对此,引入并改进残差神经网络(ResNet-50)对驾驶员分心行为进行识别。为了进一步提高模型特征的提取能力,从改进后的ResNet-50和EfficientNet-B0模型中提取特征,并将其双线性融合,从而进一步提高模型的识别准确率。通过对正常驾驶、玩手机、打电话、喝水、向后座拿东西、与副驾交谈6种驾驶员的行为进行测试,改进后ResNet-50模型的平均识别准确率达94.2%。将改进后的ResNet-50与EfficientNet-B0模型进行融合,融合模型的平均识别准确率高达96.7%。实验结果表明,该方法对驾驶员分心行为的检测有较好的分类效果。

Driver Distraction Recognition Using Bilinear Fusion Networks

Accurate recognition of driver's distraction behavior can radically reduce the traffic accidents. Traditional recognition methods have the problems of few classification categories and low accuracy. To solve these problems, residual neural network (ResNet-50) is employed to recognize driver's distraction behavior and improve the network. To further improve the feature extraction ability of the model, bilinear fusion is carried out on the features extracted from the improved ResNet-50 model and EfficientNet-B0 model. Thus, the recognition accuracy of the model can be further improved. The average accuracy of the improved ResNet-50 single model is up to 94.2%, and the average accuracy of the model after fusing the improved ResNet-50 with EfficientNet-B0 is up to 96.7%. Experimental results show that this method has a good classification effect on the detection of driver's distraction behavior.