Hear Your Vision
Canes for the blind were invented before the 1920s, it's about time they get an upgrade.
Created on 16th February 2021
•
Hear Your Vision
Canes for the blind were invented before the 1920s, it's about time they get an upgrade.
The problem Hear Your Vision solves
The lack of the ability to perceive what’s in front of you is a major problem that certain people have to face and have to rely on someone to help them in walking around. According to a recent study in 2020, around 304.1 million people globally have a form of visual impairment, out of whom 49.1 million were blind. The conventional tools for them (like a cane) have limitations in detecting obstacles since they a) have a short detection range, b) might get stuck in uneven surfaces or cracks and c) require constant effort from the user. Also, it is not very efficient in determining the shape and size of the object. Hence, we propose to build a convenient wearable device that can detect the type and proximity of the obstacle in front of the user and alerts them using sound or voice-based devices to guide them through the surroundings and avoid any collision without any extra help from other conventional aids.
THE IDEA
Our idea is to create an intelligent agent which observes the surroundings to detect objects and their proximity in real-time. The agent can be integrated with the user’s regular glasses so that it can be pointed wherever the person gazes. It will comprise 3 main modules:
- Vision-based sensor (camera)
- Vision Processing Unit (VPU, for edge computing capabilities)
- Wireless Transmission module
...
(complete idea in the attached document in links section)
Challenges we ran into
Our initial model of the agent was inspired from conventional IoT-based systems, where the sole purpose of the sensor is to capture and transmit the data to a remote server for computing. Given that we require a vision based sensor for our agent, this process would require higher bandwidth and more latency between capturing the image and alerting the user. Also, the privacy of user data might have been compromised for the want of faster processing. This led us to follow an edge-computing based approach with a suitable remote interface to receive the processed data and issue alerts.
(Challenges in prototype) The limitation of the current prototype model is that it uses ultrasonic sensors for object detection, which can only determine a range of distance between the user and the obstacle, and is thus suitable for domestic purpose only. There is no module for object identification or edge computation in the current prototype.