Assistive Vision - Android App
Overview
An Android application designed to assist visually impaired users by providing real-time currency recognition, cloth pattern identification, and text-to-speech capabilities.
Features
Currency Recognition
- Predicts Indian currency denominations with 85% accuracy
- Uses pre-trained EfficientDet-Lite0 model optimized for mobile
- TensorFlow Lite integration for on-device inference
Cloth Pattern Recognition
- Identifies cloth patterns with 76% accuracy
- Helps users distinguish between different fabric types
Accessibility Features
- Google Text-to-Speech API integration for audio feedback
- ML-Kit Object Detection for general object identification
- Handwritten Text Recognition for reading documents
Technical Implementation
- Model: EfficientDet-Lite0 fine-tuned for currency classification
- Platform: Android with TensorFlow Lite
- APIs: ML-Kit, Google Text-to-Speech
- Optimization: Model quantization for mobile deployment
Impact
- Published research in Dickensian Journal documenting the approach
- Demonstrates practical application of ML for accessibility
- End-to-end mobile ML pipeline from training to deployment
Tech Stack
TensorFlow TensorFlow Lite Android ML-Kit EfficientDet Java Python