Related Experiment Video
Updated: Aug 6, 2026
12:39
A Methodology for Capturing Joint Visual Attention Using Mobile Eye-Trackers
Published on: January 18, 2020
Driver behavior recognition under multiple illumination conditions based on attention mechanisms
Huizhi Xu1, Xinying Tao2, Yuanming Zhang2
1School of Civil Engineering and Transportation, Northeast Forestry University, Harbin, 150000, China; National Engineering Research Center of Road Safety Control Technology, China.
Summary
This study introduces a new AI model for recognizing driver actions, even in challenging lighting. The dual-stream spatiotemporal attention network (DSTA-Net) significantly improves accuracy in low-light conditions for intelligent driving systems.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Intelligent Transportation Systems
Background:
- Driver behavior recognition is crucial for intelligent driving safety.
- Complex illumination conditions pose significant challenges for existing recognition models.
- Robust spatiotemporal feature extraction is essential for accurate behavior analysis.
Purpose of the Study:
- To develop a robust driver behavior recognition model adaptable to complex illumination.
- To enhance temporal sensitivity and illumination adaptability in spatiotemporal networks.
- To provide an efficient framework for intelligent driving scenarios.
Main Methods:
- Proposed a dual-stream spatiotemporal attention network (DSTA-Net) based on the SlowFast backbone.
- Integrated Temporal-Channel Attention Module (TCAM), Temporal Attention Focusing Algorithm (TAFA), Adaptive Rank Pooling Dynamic Image Generation (ARPDIG), and Coordinate Attention (CA).
- Employed a computationally efficient (2+1)D convolutional structure for optimized spatiotemporal representation.
Main Results:
- Achieved 98.76% accuracy on the MAID-Behav dataset under normal lighting.
- Demonstrated improved performance in low-light environments, outperforming state-of-the-art models by over 7%.
- Validated the model's effectiveness in multi-view and multi-illumination scenarios.
Conclusions:
- DSTA-Net offers a robust solution for driver behavior recognition under varying illumination.
- The model's enhanced temporal and illumination adaptability contribute to superior performance.
- The proposed framework shows promise for real-world intelligent driving applications.
