Sidewalk Navigation Aid
For Visually Impaired
My Role
UX Researcher, Designer, System developer
My Team
6 Members
(3 students, 3 professors)
Duration
12 weeks
Tools
OpenCV, Python, Google Colab, Figma
Contributions
-
Spearheaded UX Research Methods and Strategies -Interviews, Secondary research and User Personas
-
Designed and Preprocessed authentic Datasets for model training
-
Conducted Model Training with 7 classifier groups
-
Facilitated Usability Testing to Evaluate the User-centered Solution using Task Success Rate Testing.
Let's get some Backstory
2022
During my undergraduate studies, I had the opportunity to collaborate on a Computer Vision project aimed at enhancing accessibility for visually impaired individuals.
Our team sought to address an underexplored challenge specific to navigation within their daily lives.
Through brainstorming sessions and in-depth user research, we identified a recurring issue they faced while navigating sidewalks and roadways – collisions with streetlights and tree trunks.
This discovery motivated us to focus our efforts on improving accessibility for the visually impaired in the context of safe roadside navigation and build a user-centered solution.
Problem Statement
How might we help in improving accessibility for the Visually Impaired while navigating on sidewalks ?
How might we prevent collisions with streetlights poles and tree trunks?
Proposed Objective
To make reliable detection and classification of streetlight poles and tree trunks
To build a Light weight, Cost effective, low power embedded system
To receive an accurate and prompt feedback mechanism
But how did we get to the Solution?
I followed a three phase Design process for defining the problem and building a solution.
Feel free to click and go to any particular section of the process
Understanding our Users
We conducted in-depth user interviews of complete or near visual impairment people from The Poona Blind Men's association
12
Visually Impaired participants
18-80
Target age group
Research goal through interviews was to understand the challenges they face on a daily basis while traversing, frequency and intensity of these pain points, their expectations, and alternatives they seek.
Insights from Interviews
All participants go for daily recreation to parks and run errands and use sidewalks for traversing
~
83%
75%
100%
~
83%
11 participants complained about having constant collisions
11 participants can't afford a smart system for navigation
9 participants specified having frequent and at times injurious collisions with
erratic Streetlight poles and tree trunks
All participants resort to walking sticks
How big is the Problem?
According to 2023 report of WHO 2.2 Billion people have visual impairment [1]
One of the most strenuous challenges faced by Visually Impaired people is the
Because of unpredictable placement of streetlight poles, open manholes, or even pedestrians
Footpaths can be one of the highly hazardous places [3]
Because of the limitations that visually impaired people face during navigation, they
Feel left out of many social activities [4]
Identifying Research Gaps
In order to understand how current technologies fail to solve this problem, we referenced through 37 research papers
Multiple ETAs (Electronic Travel Aids) in the market
Where they fail?
Multiple obstacle detection with delayed feedback
No technology solely for obstacle detection on sidewalks
Heavy and bulky systems, often uncomfortable
Expensive technology which is not affordable for general public
Meet Shyam
We created a user persona to understand the needs and pain points of a typical user of our solution
Shyam Kulkarni
Shyam has been visually impaired since the last 20 years. He lost his vision in an unfateful accident. He lives in Pune and enjoys his evening and morning walks to the park.
While walking on roadsides, I can estimate and protect myself from colliding with people. But avoiding collision with immobile obstacles like trees and poles is difficult
Pain points
-
Lack of warning prompts while walking on roads and road-sides.
-
Scared of colliding and getting hurt.
-
Trees on footpaths are sometimes at unexpected positions which leads to constant fear of collision.
-
Lack of social interaction due to fear of accidentally colliding while walking on footpaths.
-
Smart technology for navigation is expensive and unaffordable.
Needs
-
Needs help in walking freely on roadsides without colliding anywhere
-
Needs accurate estimate of obstacles to prevent accidents on footpaths.
-
Demands timely cues for getting an idea of obstacles ahead.
-
Needs to remove the fear of colliding somewhere while walking on unfamiliar roadsides.
Brainstorming Ideas
We brainstormed solution ideas to guage the system requirements and design
Developing the System
We started developing the system on Jupyter notebook using Python language
Authentic Database
9232
Images
System level block diagram
Classifiers and Accuracy
Most accurate classifier
Decision Tree
82.24%
Deployed the model on RaspberryPi microprocessor
Why RaspberryPi?
-
The RAM of RaspberryPi is proficient in Image processing and multitasking for detection and classification.
-
The microprocessor is adept at processing the large amount of data contained in our database.
Response Time through earpiece
1780ms
Usability Testing
In order to test the functionality of the system on real users and follow a user centered approach, and used the usability testing method of Task Success Rate Testing
4 participants were recruited to walk on a sidewalk with streetlight poles and tree trunks on either side
Success criteria -If the system is able to detect, classify and respond correctly
Failure criteria -
(i) If the system is unable to detect, classify or respond correctly
(ii) If the system is unable to respond
20
Experiments
85%
Success
70%
Success
77.5%
Average Success Rate
Design is all about Iterations
After the usability testing, 75% of the participants reported a feedback of lagging responses, which they expected to be faster. After multiple Brainstorming sessions, and trial and error methods we could pinpoint the problem was arising because of overfitting of the machine learning model
Next steps
Lagging response through earpiece
Brainstorming
Database Design
Model Training
Response Time through earpiece
2.78s
Response Time through earpiece
0.57s
Tested the system and got it approved by user
From this experience, I've come to understand that design is not a one-time masterpiece but a continuous process of evolution and refinement, centered around the user.
Documentation Results
We decided to document our research work and it was well-received
The research was selected for presentation at International Conference on
Ubiquitous Computing and Intelligent Information Systems [ICUIS 2022]
The paper got published in reputed International journal
Springer
Future Scope
The problem space can be broadened to accommodate potential future directions.
For improving safe navigation on sidewalks for visually impaired, the detection can be extended beyond tree-trunks and streetlights poles to other obstacles found through research like open manholes, wooden benches etc.
The system could be integrated with geospatial maps for guiding the user along with possible obstacle detection and feedback.
What I Learnt
User-centered process
I could comprehend how understanding the users and their needs proved to be a critical approach while designing technology.
Data supports UX
Backing the UX design process with data acquired from initial research as well as later user testing helps in making the iterative design pro
Designing with Computer Vision
Integrating robust machine learning models with detection using computer vision and designing a friendly Human-machine
Thanks for reading
Check out my other projects
You can find me at
If not here, then you can definitely find me petting some dog 🐶