Related Experiment Video
Updated: Oct 13, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
4.4K
A Hybrid Deep Learning Model for Recognizing Actions of Distracted Drivers
Shuang-Jian Jiao1, Lin-Yao Liu1, Qian Liu1
1Department of Civil Engineering, College of Engineering, Ocean University of China, Qingdao 266100, China.
Sensors (Basel, Switzerland)
|November 13, 2021
Summary
This study introduces a hybrid deep learning model to detect distracted driving by analyzing driver actions using skeleton data. The framework effectively identifies driver distractions, paving the way for advanced vehicle safety systems.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Automotive Safety
Background:
- Increasing in-vehicle information systems lead to more driver distractions and accidents.
- Existing computer vision methods often overlook temporal information crucial for action recognition.
- Driver distraction detection is vital for developing effective driver assistance systems.
Purpose of the Study:
- To propose a hybrid deep learning model for recognizing distracted driver actions.
- To improve the accuracy of driver distraction detection by incorporating temporal and spatial features.
- To provide a foundation for vehicle distraction warning systems.
Main Methods:
- Utilized OpenPose for extracting human skeleton information.
- Engineered artificial features (vector angles, modulus ratios) and fused them with deep network features.
- Employed K-means clustering and inter-frame comparison for keyframe selection.
- Developed a two-layer long short-term memory (LSTM) network for spatiotemporal feature extraction and a softmax layer for classification.
Main Results:
- The proposed hybrid model effectively identifies distracted driver actions.
- Fusion of artificial and deep network features enhanced spatial information density.
- The model demonstrated effectiveness on a collected dataset, validating its performance.
- The framework successfully captured crucial spatiotemporal features for accurate recognition.
Conclusions:
- The developed hybrid deep learning model is effective for distracted driver action recognition.
- Incorporating skeleton-based artificial features alongside deep learning improves detection accuracy.
- This research offers a theoretical basis for developing advanced vehicle distraction warning systems.
