Aiming at the problem that the existing visual methods are vulnerable to the interference of illumination and background in complex driving environments,a method for identifying driving distraction behaviors based on skeleton features is proposed. Thirteen key skeleton joint points of drivers are extracted through OpenPose, and multi-dimensional Action Description Features (ADFs), including skeleton structure vectors, skeleton joint vector angle vectors, and skeleton modulus ratio vectors, are constructed. A feature-selection strategy is adopted to screen out 7-dimensional significant features. On this basis, a deep residual network is improved to enhance the model's generalization ability. Experiments are carried out on public driving behavior datasets. The results show that the accuracy of the proposed method on the dataset with human semantic segmentation reaches 97.33%. Furthermore, real-world scenario validation demonstrates that the proposed method maintains good recognition stability under complex variations in illumination, background, and viewpoint. This research provides an efficient and robust technical support for vehicle monitoring systems and intelligent traffic safety early warning, and has important practical application value.
为提高网络模型的训练效率和性能,加速收敛速度,本文构建基于ResNet50的驾驶分心行为分类模型。ResNet(Residual Network)[18]通过引入残差模块,添加了“捷径连接”或“跳过连接”,允许网络训练更深结构,而不会引发梯度消失或梯度爆炸。另外,ResNet还采用了全局平均池化(Global average pooling)替代全连接层,减少了参数数量,降低了过拟合的风险,且使网络对输入图像的尺寸变化更加鲁棒。
State farm distracted driver detection是由Kaggle竞赛平台发布的驾驶行为数据集。该数据集收录了车辆行驶过程中驾驶员各类分心行为的图像,采集自26名不同体型、肤色的受试者,总计包含张图像。如图7所示,数据集涵盖9类典型分心行为及正常驾驶行为的图像样本。该数据集仅提供驾驶行为分类标签,未包含人体语义标注信息,因此通常适用于端到端驾驶行为分析任务。
ZhangLi-xia, LiuTao, PanFu-quan, et al. Analysis of effects of driver factors on road traffic accident indexes[J]. China Safety Science Journal, 2014, 24(5): 79-84.
[3]
ZhangY, ChenY, GaoC. Deep unsupervised multi-modal fusion network for detecting driver distraction[J]. Neurocomputing, 2021, 421: 26-38.
ChenRen-xiang, HuChao-chao, HuXiao-lin, et al. Driver distracted driving detection based on improved YOLOv5[J]. Journal of Jilin University (Engineering and Technology Edition), 2024, 54(4): 959-968.
[6]
BahetiB, TalbarS, GajreS. Towards computationally efficient and realtime distracted driver detection with MobileVGG network[J]. IEEE Transactions on Intelligent Vehicles, 2020, 5(4): 565-574.
ZhaoXia, LiZhao, FuRui, et al. Real-time detection of distracted driving behavior based on deep convolution-tokens dimensionality reduction optimized visual Transformer model[J]. Automotive Engineering, 2023, 45(6): 974-988.
[9]
LiuS, WangY, YuQ, et al. CEAM-YOLOv7: improved YOLOv7 based on channel expansion and attention mechanism for driver distraction behavior detection[J]. IEEE Access, 2022, 10: 129116-129124.
ChenJun, ZhangLi, ZhouBo, et al. Driver distracted driving behavior detection based on cascaded convolutional neural network[J]. Science Technology and Engineering, 2020, 20(14): 5702-5708.
[12]
PingP, HuangC, DingW, et al. Distracted driving detection based on the fusion of deep learning and causal reasoning[J]. Information Fusion, 2023, 89: 121-142.
[13]
AlkinaniM H, KhanW Z, ArshadQ, et al. HSDDD: a hybrid scheme for the detection of distracted driving through fusion of deep learning and handcrafted features[J]. Sensors, 2022, 22(5): s22051864.
YinZhi-shuai, ZhongShu, NieLin-zhen, et al. Distracted driving behavior detection based on human pose estimation[J]. China Journal of Highway and Transport, 2022, 35(6): 312-323.
HeYi, LuMan-ke, GaoSong, et al. Distracted behavior detection of commercial vehicle drivers based on the MobileViT-CA model[J]. China Journal of Highway and Transport, 2024, 37(1): 194-204.
ZhangBin, FuJun-yi, XiaJin-xiang. A metric space optimized method for driver distraction recognition model training[J]. Automotive Engineering, 2022, 44(2): 225-232.
[20]
SullmanM J M. An observational study of driver distraction in England[J]. Transportation Research Part F: Traffic Psychology and Behaviour, 2012, 15(3): 272-278.
[21]
SullmanM J M, PratF, TasciD K. A roadside study of observable driver distractions[J]. Traffic Injury Prevention, 2015, 16(6): 552-557.
[22]
PratF, PlanesM, GrasM E, et al. An observational study of driving distractions on urban roads in Spain[J]. Accident Analysis and Prevention, 2015, 74: 8-16.
[23]
KountouriotisG K, SpyridakosP, CarstenO M J, et al. Identifying cognitive distraction using steering wheel reversal rates[J]. Accident Analysis and Prevention, 2016, 96: 39-45.
[24]
CaoZ, HidalgoG, SimonT, et al. OpenPose: realtime multi-person 2D pose estimation using part affinity fields[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(1): 172-186.
[25]
HeK, ZhangX, RenS, et al. Deep residual learning for image recognition[C]∥IEEE Conference on Computer Vision and Pattern Recognition, New York, USA, 2016: 770-778.