Recognition of a Virtual Maze Scene Using Simulated Prosthetic Vision

Area of Science:

• Neuroscience and Biomedical Engineering
• Human-Computer Interaction
• Computer Vision

Background:

• Developing effective visual prosthetic systems is crucial for restoring sight to individuals with visual impairments.
• Optimizing real-time image processing and display strategies is essential for providing useful visual information through prosthetic implants.
• Simulating daily visual activities using maze pathfinding tasks can evaluate prosthetic system performance.

Purpose of the Study:

• To build a real-time psychological physics experimental platform for optimizing visual prosthetic strategies.
• To determine the best real-time image processing strategy for limited electrode counts.
• To identify the most effective visual information parameters for prosthetic implants.

Main Methods:

• Created simple (5x5), medium (8x8), and complex (11x11) 3D maze models using 3DMAX.
• Developed virtual scenes in Unity for real-time pixel processing (binarization, color inversion, template matching).
• Tested resolutions (16x16, 24x24, 32x32) and viewing angles (45°, 60°) in maze pathfinding tasks.

Main Results:

• A 45° viewing angle was significantly better than 60° for maze entrance finding and completion time.
• Increasing maze difficulty increased entrance finding time but decreased overall completion time.
• Higher resolutions (24x24, 32x32) significantly reduced pathfinding time and increased accuracy, with 32x32 achieving 100% accuracy.

Conclusions:

• The 45° viewing angle and 32x32 resolution are optimal parameters for the developed visual prosthetic system.
• The platform effectively simulates visual tasks for individuals with blindness, aiding in the development of better prosthetic vision.
• Real-time image processing strategies and resolution significantly impact the usability and effectiveness of visual prosthetics.