Research Info

Project Title

anEye: SAFE NAVIGATION IN FOOTPATH FOR VISUALLY IMPAIRED USING COMPUTER VISION TECHNIQUES

Project Type

Research Project

Research Field

Computer Vision, Machine Learning, Deep Learning

Motivation

Over 6 million people in Bangladesh need vision correction by spectacles and other means. Among them about 1.4 million children under the age of 15 are blind. Being human, we all need to move from one place to another no matter whether we are physically challenged or not. Compared to developed countries, we can easily deduce the complexity of moving around for a visually impaired in our country especially in the urban areas. The footpaths are occupied with hawkers and shops which left a little space to walk. They are narrow and often broken. Even normal people face problems while walking through the footpaths due to a huge amount of shops and hawkers. Also, there are uncovered manholes, pillars, trees, and often parked vehicles in those footpaths which leaves visually impaired peoples nearly impossible to walk on and pass by.

On the other hand, in recent years we see the enormous development in the field of artificial intelligence and computer vision. Advanced machine learning and deep learning algorithms are being applied almost in every sector. We can also observe this in the advancement of autonomous vehicles technology. Very large scale studies are being undertaken to build self-driving cars although it’s mostly for our luxury. By watching the evolving autonomous technologies we asked ourselves what if we can use some key-concepts from autonomous vehicles and use it to minimize a social problem while keeping the cost minimal? What if we can build a system that may help the blind or visually impaired in their day-to-day movement?

Therefore, to assist peoples with vision-problem we decided to come up with an efficient but feasible solution and build an intelligent system that can guide them while walking through footpaths in real-time. The everyday struggle of visually challenged people and autonomous vehicle technology motivated us to do this research.

Expected Outcomes

• An annotated footpath image dataset of Dhaka city footpaths for semantic segmentation.

• Deducing a deep learning model for footpath identification after comparative analysis to similar algorithms in terms of computational cost and accuracy.

• A low-cost unique object distance estimation model along with object detection.

• An image processing model that can analyze images and tell the user in which way they should move to avoid obstacles.

Research Questions

• How to identify safe footpath? What existing methods can be used and which one is the most suitable?

• What algorithm should be perfect for real-time obstacle detection in this scenario?

• How to estimate identified obstacle distance without using external sensors/hardware?

• How to generate suggestions for the safe pathway?

Additional Materials