Related Experiment Video
Updated: Jul 31, 2026
10:28
Dynamic Digital Biomarkers of Motor and Cognitive Function in Parkinson's Disease
Published on: July 24, 2019
LDIE-FDNet: Lightweight dynamic image enhancement-enabled real-time fatigue driving detection network
Chunyu Dong1, Tinglei Zhang1, Jing Liu1
1Xi'an Key Laboratory of Human-Machine Integration and Control Technology for Intelligent Rehabilitation, School of Computer Science, Xijing University, Xi'an, China.
Plos One
|April 1, 2026
Summary
A new lightweight network, LDIE-FDNet, improves fatigue driving detection accuracy, especially in low light. This real-time model enhances image quality and feature extraction, reducing errors and improving safety.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Automotive Safety
Background:
- Existing fatigue driving detection models struggle with accuracy-real-time balance and low-light performance.
- Inaccurate fatigue detection poses significant risks to road safety.
Purpose of the Study:
- To design a Lightweight Dynamic Image Enhancement-Enabled Real-time Fatigue Driving Detection Network (LDIE-FDNet).
- To improve accuracy and real-time performance in fatigue detection, particularly under low illumination conditions.
- To enhance feature extraction and reduce computational costs.
Main Methods:
- Image enhancement using Multi-Scale Retinex-Based Low-Light Image Enhancement Network (MSR-LIENET).
- Lightweight feature extraction via GSConv_C3k2 module with variable convolution kernels.
- Dynamic Hierarchical Feature Aggregation and Reconstruction Network (DHFAR-Net) with DySample and Semantic and Detail Infusion (SDI) for enhanced semantic and detail information.
- Multi-level feature fusion and Powerfull-IoU (PIoU) loss function for improved detection and localization.
- Integration of Maximum Closing Time (MCT) and Maximum Yawn Duration (MYD) metrics.
Main Results:
- On the YawDD dataset: mAP increased by 0.6% to 99.2%, parameters reduced by 24%, GFLOPs increased by 14.3%, and FPS increased by 23.1%.
- On the DMS dataset: mAP increased by 0.7% to 92.9%, parameters reduced by 24%, GFLOPs increased by 14.3%, and FPS increased by 20.5%.
- The model demonstrates lower reasoning delay, memory occupation, and floating-point operations.
Conclusions:
- LDIE-FDNet effectively balances accuracy and real-time performance for fatigue driving detection.
- The proposed method significantly improves detection accuracy, especially in challenging low-light conditions.
- The network achieves superior efficiency with reduced parameters and computational load.
