Related Experiment Video
Updated: Jul 8, 2026
07:15
Tactile Vibrating Toolkit and Driving Simulation Platform for Driving-Related Research
Published on: December 18, 2020
YOLO-MIRNet: detection of distracted driving with enhanced generalization using a lightweight multi-scale interaction
Chenjie Zhang1, Kunze Ba1, Xuemei Bai1
1School of Electronic Information Engineering, Changchun University of Science and Technology, Changchun, Jilin, China.
Traffic Injury Prevention
|July 1, 2026
Summary
This study introduces YOLO-MIRNet, a lightweight framework for detecting distracted driving. The model achieves high accuracy and robust generalization across diverse scenarios, enhancing road safety.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Road Safety
Background:
- Distracted driving is a major cause of road accidents.
- Current deep learning models struggle with generalization in varied driving conditions.
- Real-time driver monitoring is crucial for improving traffic safety.
Purpose of the Study:
- To develop a robust and adaptable deep learning framework for distracted driving detection.
- To enhance model generalization across different scenarios, devices, and environments.
- To provide a lightweight solution for practical, real-time traffic safety applications.
Main Methods:
- Proposed YOLO-MIRNet, a lightweight framework based on YOLOv8 architecture.
- Integrated Global and Detail Feature Aggregation (GDFA) for multi-scale feature extraction.
- Incorporated Multi-scale Adaptive Gated Fusion (MSAGF) and improved Lightweight Channel Attention (iMLCA) modules for enhanced feature discrimination and fusion.
- Trained and evaluated on the StateFarm and AUC datasets.
Main Results:
- Achieved 99.92% precision and 99.48% mAP on the StateFarm dataset.
- Attained 87.03% precision and 93.99% mAP on the AUC dataset, outperforming existing models.
- Demonstrated strong cross-dataset generalization (67.03% to 72.61%) and environmental adaptability (day/night scenarios).
Conclusions:
- YOLO-MIRNet offers an optimal balance of accuracy, lightweight design, and generalization.
- The framework provides an efficient solution for real-time distracted driving monitoring.
- This technology supports the practical deployment of intelligent traffic safety systems.
