Page 734 - AI for Good Innovate for Impact
P. 734
AI for Good Innovate for Impact
Use Case 2: AI-Driven Drowsiness Detection: Enhancing Road
Safety through Inclusive and Real-Time Monitoring
Organization: Catalink Limited
Country: Cyprus
Contact Person(s):
Konstantinos Avgerinakis, koafgeri@ catalink .eu
Zenon Lamprou, zenon.lambrou@ catalink .eu
Maria Eleni Skarkala, maria.skarkala@ catalink .eu
1 Use Case Summary Table
Item Details
Category Accessibility
Current Driver Monitoring Systems (DMS) struggle to accurately detect
drowsiness leading to false negatives and reduced reliability due to
biases in the models. These issues stem from insufficient diversity in
Problem Addressed training data and overly generalized models, resulting in biased perfor-
mance. To ensure fairness, accuracy, and inclusivity, it is essential to
address these limitations and improve the ethical deployment of AI in
driver monitoring systems.
Training unbiased and fair models to be able to detect drowsiness
Key Aspects of Solu-
tion detection when driving. This solution is primarily designed for public
transport, but it can also be utilized by individual drivers.
Drowsiness Detection, Vision Transformers, Mood Estimation, Mobile
Technology Keywords
Application, Artificial Intelligence(AI)/Machine Learning(ML)
Public
Yawning and Distracted Driver Dataset. YawDD (Yawning Detec-
Data Availability tion Dataset) [6], Driver-Drowsiness-Dataset-D3S [10], National Tsing
Hua University(NTHU) driver detection dataset [11] and private data
collected when a user uses the mobile application.
Metadata (Type of Image, video
Data)
Model Training and Train a generic VisionTransformer model on Drowsiness Detection.
Fine-Tuning Fine-tune the same models for user customization.
Testbeds or Pilot The system is currently in the research and prototype validation stage.
Deployments While a formal testbed or public pilot deployment does not yet exist.
698

