AI-Powered Drowsiness
Detection System
Thispresentation details the development of an AI-powered
drowsiness detection system. We explore the dataset used,
algorithm selection, project workflow, and results. The system
aims to improve road safety by detecting driver drowsiness in
real-time.
by Minal Fatyma
2.
Dataset Composition
Public Data
YawDD(Yawning Detection
Dataset) provides a foundational
dataset.
Private Data
Self-collected video samples
augment the dataset with diverse
scenarios.
Dataset Size
107 videos (61 drowsy, 57 non-
drowsy), 4,202 frames (~300
frames per video).
3.
Algorithm Selection
1 SVM
Achieved90.97% accuracy.
2 LSTM
Reached 93.59% accuracy.
3 MobileNetV2
Outperformed others with 100% accuracy and real-time
performance.
4 InceptionV3
Less than 80% (Needs Improvement)
4.
Project Workflow
Data Collection
Gatherpublic and private video samples.
Preprocessing
Extract frames and detect facial landmarks.
Feature Extraction
Identify signs of drowsiness.
Model Training
Train CNN/LSTM models for classification.
Real-Time Detection
Process live video and generate alerts.
Alert Mechanism
Timer-based alert system.
Monthly Timeline
1
Month 1
Literaturereview and data collection.
2 Month 2
Data preprocessing and model selection.
3
Month 3
Model training and testing.
4 Month 4
Real-time integration and evaluation.
5
Month 5
System optimization and report writing.
7.
Results & Evaluation
ModelPerformance
Analyzed with accuracy, confusion matrix, and metrics.
Real-World Testing
System tested in diverse lighting and driving conditions.
8.
Future Work
Improve Accuracy
Enhanceperformance under various lighting conditions.
Vehicle Integration
Integrate with vehicle control systems for automated safety.