This project is based on the RBI's tender titled "Development of Mobile Application for Aiding the Visually Impaired in Identifying the Denominations of Indian Banknotes (Currency Denomination Identifier)" which was released on May 04, 2019.
This android app will help visual and hearing-impaired persons to detect the Indian Currency Denomination by placing a banknote under the camera.
This is a real-world project which is based on (Research + Applied) machine learning and Android Development.
We created the Indian Currency Dataset(ICD) for this project which currently has 12,000 images. ICD is a very rich dataset, which contains very diverse images. We have covered all the classes (none, 5, 10, 20, 50, 100, 200, 500, 2000). Also, we have covered all kind of note's images like older and newer, front and rear, in different lighting conditions etc.
(We will share the dataset after the completion of the project.)
We are using Keras, TensorFlow, Fastai and PyTorch libraries, to leverage the power of all.
Currently, we have achieved 99% accuracy on ResNet50 and 96% accuracy on MobileNetV2 model (DATE : 28/9/2019).
We are using MobileNet model because it runs efficiently and fast on mobile phones.
[DATE : 12/9/2019]