Vision Transformer Customized for Environment Detection and Collision Prediction to Assist the Visually Impaired

Area of Science:

• Computer Vision
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Visually impaired individuals face significant challenges in independent navigation.
• Existing assistive technologies often lack real-time environmental awareness and intuitive feedback mechanisms.

Purpose of the Study:

• To develop and evaluate a novel navigation and collision avoidance system for the visually impaired.
• To leverage vision transformers and multimodal feedback for enhanced situational awareness.

Main Methods:

• Implementation of vision transformers for accurate, real-time object detection.
• Development of algorithms for semantic segmentation and trajectory vector generation of detected objects.
• Integration of audio and vibrotactile feedback modules for multimodal collision warnings.

Main Results:

• The system achieved 95% accuracy in object classification across diverse indoor and outdoor conditions.
• Demonstrated capability for real-time identification and trajectory prediction of potential obstacles.
• Successfully integrated multimodal feedback for conveying collision risks.

Conclusions:

• The proposed system, integrating vision transformers and multimodal feedback, shows significant promise for practical navigation assistance.
• The technology offers a reliable and feasible solution to improve the independence and safety of visually impaired users.
• Further experimental validation will assess the system's usability and efficiency in real-world scenarios.