基于双分支和可变形卷积网络的驾驶员行为识别方法  

Driver behavior recognition method based on dual-branch and deformable convolutional neural networks

作　　者：胡宏宇[1] 张争光 曲优 蔡沐雨 高菲[1] 高镇海[1] HU Hong-yu;ZHANG Zheng-guang;QU You;CAI Mu-yu;GAO Fei;GAO Zhen-hai(State Key Laboratory of Automotive Simulation and Control,Jilin University,Changchun 130022,China)

机构地区：[1]吉林大学汽车仿真与控制国家重点实验室,长春130022

出　　处：《吉林大学学报(工学版)》2025年第1期93-104,共12页Journal of Jilin University:Engineering and Technology Edition

基　　金：吉林省重大科技专项项目(20230301008ZD);吉林省自然科学基金项目(20210101064JC);国家自然科学基金项目(52272417,52202495)。

摘　　要：针对汽车座舱内的驾驶员行为识别任务,本文提出了一种基于双分支神经网络的识别方法。网络模型的主分支以ResNet 50作为主干网络进行特征提取,利用可变形卷积使模型适应驾驶员在图像中的形状和位置变化。辅助分支在梯度反向传播过程中辅助更新主干网络的参数,使主干网络能够更好地提取有利于驾驶员行为识别的特征,从而提高模型的识别性能。网络模型在State Farm公开数据集的消融实验和对比实验结果表明:本文网络模型的识别准确率可以达到96.23%,针对易于混淆的行为类别识别效果更佳。研究结果对于汽车座舱内的驾驶员行为理解与保障行车安全具有重要意义。This research offers a recognition approach based on a dual-branch neural network for recognizing driver behavior in the vehicle cockpit.The main branch of the network model employs Res Net50 as the backbone network for feature extraction,and employs deformable convolution to adapt the model to changes in the shape and position of the driver in the image.The auxiliary branch aids in updating the parameters of the backbone network during the gradient backpropagation process,so that the backbone network can better extract features that are beneficial to driver behavior recognition,thereby improving the recognition performance.The results of ablation experiments and comparing experiments of the network model on the State Farm public dataset reveal that the proposed network model has a recognition accuracy of 96.23%and a better recognition effect on easily confused behavior categories.The study′s findings are critical for understanding driver behavior in the vehicle cockpit and guaranteeing driving safety.

关 键 词：车辆工程 智能驾驶 驾驶员行为识别 卷积神经网络 辅助分支 可变形卷积 

分 类 号：U471.3[机械工程—车辆工程]