Currency Classifier Model on GitHub

Overview

A-EYE is an assistive wearable technology system engineered to empower visually impaired individuals with greater environmental awareness and independence. The system connects custom-built smart glasses equipped with a camera to an intuitive companion mobile application, delivering real-time currency denomination identification, optical character recognition (OCR), and audio feedback.

Key System Components

Smart Glasses & Edge Hardware

  • Built around the ESP32 microcontroller and embedded camera module, programmed in Embedded C.
  • Captures image frames upon user prompt and streams them efficiently over low-latency wireless communication to the companion smartphone.
  • Designed with low power consumption and compact ergonomics for comfortable daily wear.

Companion Mobile Application

  • Developed with Flutter to deliver a responsive, accessible cross-platform mobile experience.
  • Incorporates accessible voice controls, touch gestures, and instant text-to-speech audio narration to communicate findings back to the user seamlessly.
  • Handles edge image preprocessing and runs offline machine learning inference.

Deep Learning & Computer Vision

  • Currency Recognition: Trained a lightweight MobileNet Convolutional Neural Network (CNN) specifically tailored to classify Indian rupee denominations across diverse lighting conditions, backgrounds, angles, and folding states.
  • Quantized and converted the deep learning model to TensorFlow Lite (TFLite) to achieve sub-second, on-device inference without requiring active internet connectivity.
  • Text Recognition (OCR): Integrated text extraction and parsing pipelines to identify and read street signs, documents, and product packaging aloud.

Tech Stack

  • Mobile Application: Flutter, Dart
  • Hardware & Firmware: ESP32, Embedded C
  • Machine Learning: Python, TensorFlow, TensorFlow Lite, MobileNet CNN, OpenCV