This project explores a lightweight deep learning approach for driver facial expression recognition (FER) using video-based temporal context instead of only static images.
The goal is to improve emotion recognition robustness for intelligent driving and driver monitoring systems.
- Lightweight FER backbone for efficient inference
- Video-sequence modeling to capture temporal expression changes
- Focus on practical driver-state understanding for ADAS-style scenarios
- Training and experimentation assets included in
Research_Implementation/
The comparison below shows the motivation for moving from single-frame FER to sequence-aware FER:
The pipeline used in this work is summarized below:
Training behavior and model learning progress:
.
├── README.md
├── StaticvsVideo.png
├── methodology2.png
├── Training-Graph.png
├── A Novel Lightweight Deep Learning Approach for Drivers’ Facial Expression Detection.pdf
└── Research_Implementation/
├── FED-Driver.ipynb
├── First perfect test.ipynb
├── researchmodule-trial1.py
├── KMU_FED_DATASET/
└── kmu_fed/
- Main experimental code is in:
Research_Implementation/researchmodule-trial1.pyResearch_Implementation/FED-Driver.ipynb
- The implementation includes a lightweight convolutional design and attention modules for FER.
- Dataset folders under
Research_Implementation/are used for training/validation workflows.
- Open the notebooks or Python script in
Research_Implementation/. - Install required dependencies (PyTorch, TorchVision, OpenCV, scikit-learn, matplotlib, Pillow, tqdm).
- Configure dataset paths according to your local environment.
- Run training and evaluation cells/scripts.
If you use this repository, please cite the related manuscript included in this project:
A Novel Lightweight Deep Learning Approach for Drivers’ Facial Expression Detection.pdf