AccessLens 2(AN)K.AI

Inspiration

Over 285 million people are visually impaired. They take photos daily but can't see what's inside them. I wanted to build a free, offline, multi-language tool to help them understand their surroundings.

What It Does

AccessLens 2(AN)K.AI is an Image AI accessibility tool. Upload any photo and get:

• 🔍 Object detection (80+ types via YOLOv8)
• 🚨 Smart Obstacle Alerts (warns about nearby cars, people, objects)
• ♿ Accessibility Score (0-100 scene rating)
• 🌐 Multi-language (English + Urdu)
• 🔊 Voice output for blind users
• 📥 Downloadable report

How We Built It

Built with YOLOv8 (6MB, offline), Streamlit (UI), OpenCV (image processing), and gTTS (voice). Custom logic analyzes bounding box sizes to estimate object proximity — no depth camera needed. Full Urdu translation for 80+ objects.

Challenges

• Limited disk space (2.73GB) — removed 990MB model, replaced with custom logic
• Windows Smart App Control blocked pip — used python -m pip
• Built Urdu dictionary for 80+ objects from scratch
• Fixed OpenCV libGL error on Streamlit Cloud

Accomplishments

• Complete accessibility tool in under 20MB
• Under 5 seconds per image
• English + Urdu support
• Custom obstacle alerts without depth sensors
• Deployed live on Streamlit Cloud

What We Learned

YOLOv8 integration, bounding box proximity estimation, Streamlit session state, accessibility-first design, working with hardware constraints.

What's Next

Real-time camera feed, mobile app, more languages (Arabic, Hindi, Spanish), smart glasses integration.

Built With

• gtts
• image-ai
• machine-learning
• pandas
• text-to-speech