Project Conclusion
Project Evolution
- •Initial unified approach with single AI model strategy
- •Strategic pivot to multi-component specialized architecture
- •Enhanced accuracy through focused implementations
- •Development of component-specific solutions for ATM and navigation
Key Achievements
- •Successful integration of VLMs in real-time assistive systems
- •Development of specialized computer vision models for specific tasks
- •Creation of a comprehensive campus-specific dataset with 4,202 images
- •Implementation of voice-interactive interface for seamless module switching
Technical Milestones
89.4%
ATM Button Detection
Accuracy in button recognition
75
Processing Speed
Images processed in ~14 seconds
4,202
Dataset Size
Campus-specific training images
100%
Module Integration
Core components integrated
Core Components
ATM Assistant Module
- •Finger position tracking and feedback
- •YOLOv8-based button detection
- •EasyOCR for screen text recognition
- •Voice-guided interface
Campus Navigation Module
- •Custom-trained YOLOv8 model
- •Real-time hazard proximity alerts
- •Campus-specific dataset utilization
- •Voice-based interaction system
Future Impact
- •Enhanced independence for visually impaired students
- •Improved accessibility in educational institutions
- •Potential for adaptation in other environments
- •Contribution to assistive technology research