Overview
Blind Trust is an intelligent assistive system designed to help
visually impaired individuals independently identify Sri Lankan currency,
detect counterfeit banknotes, and calculate the total value of multiple notes.
The system combines IoT hardware with deep learning-based computer vision,
utilizing an Arduino Uno, camera module, OpenCV, TensorFlow, and multiple
convolutional neural networks to provide accurate real-time voice feedback.
Unlike conventional currency recognition applications, Blind Trust performs an
end-to-end workflow consisting of currency detection, denomination recognition,
counterfeit verification using UV watermark analysis, coin identification, and
automatic currency summation. The solution was developed to improve both
financial accessibility and security while providing a portable, user-friendly
interface for everyday use.
Key Features
- AI-powered IoT assistive device for visually impaired users.
- Automatic banknote detection before recognition.
- Deep learning-based denomination recognition for Sri Lankan currency notes and coins.
- Counterfeit detection using UV watermark analysis and CNN models.
- Voice-based feedback enabling hands-free interaction.
- Automatic summation of multiple genuine banknotes.
- Custom datasets collected and augmented under diverse lighting conditions, backgrounds, and viewing angles.
Technical Contributions
- Designed an end-to-end computer vision pipeline using OpenCV and TensorFlow.
- Developed custom CNN models for currency detection.
- Fine-tuned InceptionV3 for banknote and coin denomination recognition.
- Fine-tuned VGG16 for counterfeit currency classification.
- Applied preprocessing techniques including background removal, image resizing, dataset cleaning, brightness and contrast adjustment, Gaussian blur, noise injection, perspective transformation, and extensive data augmentation.
- Integrated all models into a portable Arduino-powered embedded system with real-time camera inference.
Performance
- 99% banknote denomination recognition accuracy.
- 93% coin denomination recognition accuracy.
- 97% counterfeit currency detection accuracy.
- 94% currency presence detection accuracy.
Gallery
Presentation & Videos