Driving Behavior Recognition Based on Human Feature Point Extraction and Multidimensional Time Series Classification

摘要: 驾驶人行为识别对提高驾驶安全性和发展智能交通至关重要。由于不同驾驶环境和驾驶人特征存在较大差异,基于端到端深度学习的驾驶行为识别模型很难在不同数据集下保持较高的泛化性能。针对上述问题,提出基于人体特征点提取和多维时间序列分类的驾驶人行为识别方法。使用YOLOv8和人体姿态估计器蒸馏(Distillation for whole-body pose estimators,DWPose)跟踪驾驶人区域并提取驾驶人人体特征点矩阵。对特征点矩阵进行归一化、平滑和维度转换。分别建立了基于Transformer、Informer、时间卷积神经网络(Temporal convolutional network,TCN)和注意力机制-长短期记忆网络(Attention based long short-term memory,LSTM-Attention)的多维时间序列分类模型。结果表明,Informer模型的识别准确性最高,TCN模型的运行效率最高。当使用Driver-100-Day进行训练时,Informer在Driver-100-Day、Driver-100-Night和State Farm Driver 2数据集上的测试准确度分别为90.82%、88.77%和80.67%,相比于CNN-Transformer提高了24.56%、72.02%和67.57%。所提方法相比于基于单帧图像输入的模型在泛化性能方面有较大的改善,且能够达到较高的识别效率和准确度。

Institutions

Publication Details

Journal
Journal of Mechanical Engineering
Published
2025-01-01
DOI
https://doi.org/10.3901/jme.2025.15.233
Primary Topic
Human Pose and Action Recognition
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Driving Behavior Recognition Based on Human Feature Point Extraction and Multidimensional Time Series Classification

Journal of Mechanical Engineering
Human Pose and Action Recognition
article

Driving Behavior Recognition Based on Human Feature Point Extraction and Multidimensional Time Series Classification

article en

Abstract

摘要: 驾驶人行为识别对提高驾驶安全性和发展智能交通至关重要。由于不同驾驶环境和驾驶人特征存在较大差异,基于端到端深度学习的驾驶行为识别模型很难在不同数据集下保持较高的泛化性能。针对上述问题,提出基于人体特征点提取和多维时间序列分类的驾驶人行为识别方法。使用YOLOv8和人体姿态估计器蒸馏(Distillation for whole-body pose estimators,DWPose)跟踪驾驶人区域并提取驾驶人人体特征点矩阵。对特征点矩阵进行归一化、平滑和维度转换。分别建立了基于Transformer、Informer、时间卷积神经网络(Temporal convolutional network,TCN)和注意力机制-长短期记忆网络(Attention based long short-term memory,LSTM-Attention)的多维时间序列分类模型。结果表明,Informer模型的识别准确性最高,TCN模型的运行效率最高。当使用Driver-100-Day进行训练时,Informer在Driver-100-Day、Driver-100-Night和State Farm Driver 2数据集上的测试准确度分别为90.82%、88.77%和80.67%,相比于CNN-Transformer提高了24.56%、72.02%和67.57%。所提方法相比于基于单帧图像输入的模型在泛化性能方面有较大的改善,且能够达到较高的识别效率和准确度。

Journal of Mechanical EngineeringVol. 61(15)
Chang'an University (CN)
Openalex Percentile: Top 57%
Human Pose and Action Recognition
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.