Cash Recognition for Visually Impaired

Powered by deep learning technology, a mobile app for visually impaired - that can recognize Nepalese bank notes using smartphone camera. ...learn more

Project status: Under Development

Mobile, Internet of Things, Artificial Intelligence, PC Skills

Groups
Student Developers for AI, Movidius™ Neural Compute Group, DeepLearning

Intel Technologies
AI DevCloud / Xeon, Intel Opt ML/DL Framework, Movidius NCS

Code Samples [1]Links [2]

Overview / Usage

The Nepalese currency does not have any special feature in them, so that the visually impaired can differentiate between the notes of different values. The visually impaired people, unlike us, face difficulties in day to day monetary transactions since they cannot recognize the bank notes as easily as we can in Nepal. As an initiative for the Nepalese visually impaired community, this app will aid them to recognize notes without any hassle. By hovering a smartphone over a note, this app will recognize it and play an audio enabling the user to hear and know the value of the note. The app will be bilingual, with Nepalese and English languages as audio playback options for a better user experience.

Intel Blog Article:
https://software.intel.com/en-us/blogs/2017/11/21/cash-recognition-for-the-visually-impaired-using-deep-learning

Article on Practical lessons learned while implementing it:
https://medium.com/deep-learning-journals/practical-lessons-learned-while-implementing-image-classifier-6dc39c6efd7e

Github Link for first prototype source code along with Training data and pre-trained model
https://github.com/devSessions/crvi

Methodology / Approach

Methodology is broken down into 3 separate processes, one is data collection, second is model creation and verification and third to mobile app development. The first and second processes are inter-related, as I add data of new classes, i create separate models to test its accuracy. For each new model creation i test it with the prototype app i have created. App development parts has sub sections in it. There are many features to add in app itself, so broken down into 3 months of project time, each month one major milestone in app development is achieved and the end of the third month final model is used with final cross platform app that gets build.

Technologies Used

Intel Optimized Tensorflow/Keras, React Native (Android + iOS), Python Flask server, Intel DevCloud for training, Intel Movidius Neural Compute Stick, Intel NUC kit.

Repository

https://github.com/devSessions/crvi

Collaborators

There are no people to show.

Comments (1)