Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Federated Learning for Drowsiness Detection in Connected Vehicles

William Lindskog, Valentin Spannagl, Christian Prehofer

TL;DR

The paper tackles privacy and data-sharing challenges in driver drowsiness detection by applying federated learning across a vehicular network using the YawDD dataset. It compares frame-level 2D-CNN and sequence-based 3D-CNN approaches within a Flower/PyTorch FL pipeline, reporting up to $99.2\%$ accuracy for frame-based methods and $90.1\%$ for 3D-CNN on a 3-class task (normal, talking, yawning). Key contributions include a FL framework tailored for driver monitoring, an analysis of how model performance scales with the number of federated clients, and actionable guidance on hyperparameters and architecture choice. The findings demonstrate FL’s promise for privacy-preserving, distributed drowsiness detection in automotive settings, with implications for real-world deployment and future personalization research.

Abstract

Ensuring driver readiness poses challenges, yet driver monitoring systems can assist in determining the driver's state. By observing visual cues, such systems recognize various behaviors and associate them with specific conditions. For instance, yawning or eye blinking can indicate driver drowsiness. Consequently, an abundance of distributed data is generated for driver monitoring. Employing machine learning techniques, such as driver drowsiness detection, presents a potential solution. However, transmitting the data to a central machine for model training is impractical due to the large data size and privacy concerns. Conversely, training on a single vehicle would limit the available data and likely result in inferior performance. To address these issues, we propose a federated learning framework for drowsiness detection within a vehicular network, leveraging the YawDD dataset. Our approach achieves an accuracy of 99.2%, demonstrating its promise and comparability to conventional deep learning techniques. Lastly, we show how our model scales using various number of federated clients

Federated Learning for Drowsiness Detection in Connected Vehicles

TL;DR

The paper tackles privacy and data-sharing challenges in driver drowsiness detection by applying federated learning across a vehicular network using the YawDD dataset. It compares frame-level 2D-CNN and sequence-based 3D-CNN approaches within a Flower/PyTorch FL pipeline, reporting up to accuracy for frame-based methods and for 3D-CNN on a 3-class task (normal, talking, yawning). Key contributions include a FL framework tailored for driver monitoring, an analysis of how model performance scales with the number of federated clients, and actionable guidance on hyperparameters and architecture choice. The findings demonstrate FL’s promise for privacy-preserving, distributed drowsiness detection in automotive settings, with implications for real-world deployment and future personalization research.

Abstract

Ensuring driver readiness poses challenges, yet driver monitoring systems can assist in determining the driver's state. By observing visual cues, such systems recognize various behaviors and associate them with specific conditions. For instance, yawning or eye blinking can indicate driver drowsiness. Consequently, an abundance of distributed data is generated for driver monitoring. Employing machine learning techniques, such as driver drowsiness detection, presents a potential solution. However, transmitting the data to a central machine for model training is impractical due to the large data size and privacy concerns. Conversely, training on a single vehicle would limit the available data and likely result in inferior performance. To address these issues, we propose a federated learning framework for drowsiness detection within a vehicular network, leveraging the YawDD dataset. Our approach achieves an accuracy of 99.2%, demonstrating its promise and comparability to conventional deep learning techniques. Lastly, we show how our model scales using various number of federated clients
Paper Structure (12 sections, 8 figures, 1 table)

This paper contains 12 sections, 8 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. YawDD Dataset
  4. Data Preprocessing
  5. Method for Driver Drowsiness
  6. Deep Learning and Spatial-Temporal Data
  7. Federated Learning Tools
  8. Evaluation and Results
  9. 3D-CNN and Video Sequence Processing
  10. 2D-CNN and Image Processing
  11. Discussion
  12. Conclusion

Figures (8)

  • Figure 1: Samples from the YawDD dataset in two perspectives. Top: Rear mirror. Bottom: Dash abtahi2014yawdd
  • Figure 2: Sample includes two categories.
  • Figure 3: Flower core framework architecture beutel2020flower.
  • Figure 4: 3D-CNN architecture as in ed2020real
  • Figure 5: PyTorch code for convolutional neural network we use. The parameter values for the optimizer are initial values.
  • ...and 3 more figures