Sight for the Blind

The aim for this project is to provide automation for the blind.

Blind people don't want to feel any different from others, but they are constraint from doing certain things like writing, eating on their own, walking.
This things require sight so with the use of Machine Learning and voice automation in android, we would give sight to the blind.

Insipration

While preparing for 2nd semester exams, on my way to the Library on the last week of December 2019, I meet a blind student who was himself making His way to the library,we walked and talked together, and I could only imagine how much strength and determination he puts in day after day to be a student like studying, going for classes, submitting assignment,... and doing what students do every day. Believe me its stressful
As the conversation progressed I could see He struggles, what ordinary what have been easily done by mewould take him time to get done.
So after our brief conversation, I sat him down at the library desk and setup his laptop, earphone and charger,.. but before leaving I noticed that he used microsoft-word TextToSpeech to read (that means technology can be used to solve this problem) And I was determined to use technology to help him and other visually diasabled persons(blind) see the world and the way for me to do this, is to use Machine learning.

Technology stacks

• Keras with Tensorflow 2.0 backend
• Google colab
• Android studio

How it works

By clicking on the app screen, the app would speak out what it detects, either to read lables or detect item.
The app will consist of two screens;

• Tell me
  For this section, the app will be able to read every-day items like product label, book name, food item name ...
  This will be done by point the phone directly in front of what to read
• Detect traffic
  Just by my experince, they navigate everyday and crossing the road without know the traffic sign is not such a good idea
  This setion is to detect the trafiic sign and say what the model detect, so that He can cross safely.

Featues

• On device Machine learning
• Privacy in users data

Presently

• Optical character recognotion has been implemented for Tell me section.
• Pre-trained TFLite model on 80 classes is presently used for testing the app
• Android in built TextToSpeech is in use to speak the text

Next

• Improve the OCR
• Gather data and train the model
• Convert to TFLite
• Improve the speech reconition
• Proved Real-time description of the environment for easy automation