.png){kind=icon}
.png){kind=icon}
Return
Driving skill classification via multidimensional data fusion in complex scenarios: a convolutional neural network–long short-term memory (CNN-LSTM) approach
.png){kind=icon}
OA
AI
S
Y
C
Z
DOI：10.1016/j.ijtst.2025.10.011.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate evaluation of driving skill levels is essential for optimizing personalized settings in driver assistance systems (DAS) and training programs. However, current research faces two limitations: (1) reliance on unidimensional data (maneuvers or cognition), neglecting multidimensional fusion for comprehensive evaluation, and (2) assessment in simplified scenarios, failing to evaluate critical real-world skills like interacting with merging traffic. The objective of this study is to integrate complicated driving environments and a multidimensional data fusion model, so as to classify driving skills more precisely. Data was collected from a driving simulator experiment, which included a regular scenario and three work zone scenarios with different complexities. 32 participants were equipped with psycho-physiological sensors during the experiment and completed a post-experiment survey. Driver maneuvers, vehicle responses, and cognitive features were processed in an attentional convolutional neural network–long short-term memory (CNN-LSTM) classifier, with three subjective skill levels (novice, typical, expert) derived from a post-simulator survey as training labels. The accuracy of the data fusion model was assessed and compared across individual dimensions (maneuvers, vehicle responses, and cognition), as well as different scenario complexities. Results indicate the multidimensional fusion CNN-LSTM reached a maximum accuracy of 97.49%, outperforming single-dimensional classifications based on maneuvers, vehicle responses, and cognitive features by 3.1%, 11.66%, and 23.2%, respectively. For the three scenarios, the mean accuracy was 99.21%, exceeding the performance of common road scenarios by 2.99%. These findings suggest that utilizing multidimensional data and incorporating the complexities of scenarios could significantly improve driving skill classification accuracy. This research enables real-time driver skill classification, allowing DAS interventions to adjust dynamically by skill level. It also provides quantifiable standards to tailor training content, improving novice readiness in these environments.
Keywords:
Driving skill classification
Scenario complexity
Multidimensional fusion
Attentional convolutional neural network-long short-term memory (CNN-LSTM)
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
4.8
Papers:
1.5K
Citations:
1.5K
Organization
Cited Papers