The problem trinetra solves

Problem

Visually impaired people face many challenges in their day to day life which are very convenient for others. As they might say, sidewalks can be the most dangerous of places. One of the consequences of vision loss is being uncomfortable about safety while moving around or travelling independently. Safe navigation on sidewalks is the most important requirement.

Solution Proposed

In order to solve this, we can develop hardware as well as a software solution.

• Trinetra headband: Raspberry Pi embedded with Tensorflow to detect objects in the surroundings.
• A simple Web Application which uses the device in your hand, a smartphone to read your surroundings for you. We implemented it just in case if people somehow can’t access our hardware.

Tech Stack Used

We will be using, OR have used:

• Raspberry Pi 4B module for ML Model training and deployment
• Pi Cam is a feasible option to capture images
• Lipo battery is lightweight and easily chargeable

For Web-App

• ReactJS
• TensorFlow 2.0
• react-WebCam
• CocoSSD

Challenges we ran into

Developing a working prototype of an object detection model was not a big task initially, maybe took around 4-5 hrs, but the challenge was to improve it with every use. We had to set an optimum time period for detection which will enable the user to somewhat communicate with the device. We set the threshold to 2500 milliseconds after experimenting with so many values above and beyond 0-10,000.
Assembling a CAD model with so many components which are highly relevant but also heavy in nature was a task the designer needed to accomplish.
We experimented with stuff, like using a woman's voice when operated the web-app from the phone and a man's voice when operated from a bigger screen which was fun to learn.
I had used to react JS for the first time but had experience with TF2.0 in a Python environment. Pivoting was difficult indeed.

However, my team has overcome all the challenges we ran into, emerging victorious in our own way!