基于车载传感器时空图卷积的驾驶行为识别  

Driving behavior recognition based on spatio-temporal graph convolution of vehicle sensor

作　　者：吴东昊 马春梅[1] 武盼盼 孙华志[1] 王佳宇 房一阁 WU Donghao;MA Chunmei;WU Panpan;SUN Huazhi;WANG Jiayu;FANG Yige(College of Computer and Information Engineering,Tianjin Normal University,Tianjin 300387,China)

机构地区：[1]天津师范大学计算机与信息工程学院,天津300387

出　　处：《天津师范大学学报（自然科学版）》2023年第2期74-80,共7页Journal of Tianjin Normal University：Natural Science Edition

基　　金：天津市教委科研计划资助项目(2018KJ155).

摘　　要：本文从图(graph)的角度出发,提出基于车载传感器时空图卷积的驾驶行为识别模型.首先,通过挖掘传感器间的关联性构建传感器的图结构;其次,基于时间信息融合策略的不同,提出基于LSTM的层级时空图卷积网络(H-STGCN)和修正的时空图卷积网络(M-STGCN),用于捕捉传感器的时空相关性进行驾驶行为识别;最后,在公开的2个驾驶行为数据集上进行实验,结果表明H-STGCN模型的识别效果优于现有方法.A driving behavior recognition model based on spatio-temporal graph convolution of vehicle sensors is proposed from the perspective of graph.Firstly,graph structure representation of sensors is constructed by mining the relevance of the vehicle sensors.Secondly,based on the different strategies of the time information fusion,double layer LSTM based hierarchical spatio-temporal graph convolutional network(H-STGCN)and modified spatio-temporal graph convolutional network(M-STGCN)are proposed,which can capture the temporal and spatial relevance of sensors for the driving behavior recognition.Finally,the experiment is conducted on two currently published driving behavior data sets,and the results show that the recognition effect of H-STGCN model is better than the existing methods.

关 键 词：驾驶行为识别 车载传感器 图卷积 空间依赖性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]