Related Experiment Video
Updated: May 15, 2025
08:45
A Dual Task Procedure Combined with Rapid Serial Visual Presentation to Test Attentional Blink for Nontargets
Published on: December 5, 2014
9.1K
RMTSE: A Spatial-Channel Dual Attention Network for Driver Distraction Recognition
Junyi He1, Chang Li2, Yang Xie2
1Faculty of Information Science, Huxi Campus, Chongqing University, Chongqing 400044, China.
Sensors (Basel, Switzerland)
|May 14, 2025
Summary
This study introduces a new hybrid attention model (RMTSE) to accurately recognize driver distraction, improving road safety. The model enhances feature extraction and generalization, achieving high accuracy in recognizing localized and subtle driving behaviors.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Road Safety
Background:
- Driver distraction is a major cause of traffic accidents.
- Current behavior recognition methods lack accuracy for localized actions and distinguishing subtle differences.
- Accurate driver behavior recognition is crucial for enhancing road safety.
Purpose of the Study:
- To propose RMTSE, a hybrid attention model, to improve driver distraction recognition.
- To enhance the accuracy and efficiency of recognizing localized and subtle driving behaviors.
- To address limitations in existing driver behavior recognition systems.
Main Methods:
- Developed a hybrid attention model named RMTSE.
- Introduced the Manhattan Self-Attention Squeeze-and-Excitation (MaSA-SE) module, combining spatial self-attention and channel attention.
- Employed a transfer learning strategy with pre-trained weights for accelerated convergence and enhanced feature generalization.
Main Results:
- Achieved Top-1 accuracies of 99.82% on the SFD3 dataset and 94.95% on the 100-Driver dataset.
- Demonstrated improved learning efficiency through focused feature extraction.
- Outperformed existing state-of-the-art methods with minimal parameter increments.
Conclusions:
- The RMTSE model significantly enhances driver distraction recognition accuracy.
- The MaSA-SE module effectively improves feature discriminability and learning efficiency.
- The proposed methods contribute to advancing road safety through improved driver behavior analysis.
