Simulated Prosthetic Vision: The Benefits of Computer-Based Object Recognition and Localization

Area of Science:

• Biomedical Engineering
• Neuroscience
• Computer Vision

Background:

• Current visual neuroprostheses offer limited vision, making simple tasks challenging for blind patients.
• Simulated prosthetic vision (SPV) is crucial for evaluating future neuroprosthesis capabilities.
• Previous SPV studies did not utilize advanced image processing techniques.

Purpose of the Study:

• To investigate the potential of advanced image processing, specifically object localization, to enhance visual neuroprosthesis functionality.
• To evaluate the usability of a simulated prosthetic vision device for object recognition, localization, and reaching tasks.
• To determine the effectiveness of high-level information extraction for restoring visuomotor behaviors.

Main Methods:

• Simulated a prosthetic vision device employing object localization in visual scenes.
• Assessed the device's performance in object recognition, localization, and visually guided reaching.
• Compared performance metrics (accuracy, speed) using different electrode counts (9 vs. 100).

Main Results:

• A low number of electrodes (e.g., nine) proved sufficient for restoring visually guided reaching with reasonable timing and high accuracy.
• Performance in both accuracy and speed was comparable between simulations using 9 and 100 electrodes.
• High-level information extraction (object recognition and localization) significantly enhanced simulated prosthetic vision usability.

Conclusions:

• Advanced image processing, particularly object localization, can drastically improve the functionality of current low-resolution visual neuroprostheses.
• This object localization method shows promise for restoring various visuomotor behaviors in patients with visual impairments.
• The reliability of vision algorithms remains a key limitation, though rapid advancements are expected.