Human faces detection and localization with simulated prosthetic vision

Abstract

Clinical trials reveal that current visual neuroprosthesis are not yet usable. The main reason is the small number of implanted electrodes, leading to a very poor visual resolution. The resolution is especially not sufficient to detect specific objects (faces, signs, etc.) in the surroundings. We used simulated prosthetic vision (SPV) to show that pre-processing of the camera image could restore these functions, even with low- resolution implants. Specifically, we showed that it is possible to quickly detect and localize human faces located nearby. We suggest that high-level processing of the video stream may be included in current visual neuroprosthesis. This would restore many visuomotor behaviors such as grasping, heading, steering, etc.