ARTIFICIAL VISION FOR THE BLIND: A BIO-INSPIRED ALGORITHM FOR OBJECTS AND OBSTACLES DETECTION

Abstract

Although artificial vision systems could potentially provide very useful input to assistive devices for blind people, such devices are rarely used outside of laboratory experiments. Many current systems attempt to reproduce the visual image via an alternative sensory modality (often auditory or somatosensory), but this dominant "scoreboard" approach, is often difficult to interpret for the user. Here, we propose to offload the recognition problem onto a separate image processing system that then provides the user with just the essential information about the location of objects in the surrounding environment. Specifically, we show that a bio-inspired image processing algorithm (SpikeNet) can not only robustly, precisely, and rapidly recognize and locate key objects in the image, but also in space if the objects are in a stereoscopic field of view. In addition, the bio-inspired algorithm allows real-time calculation of optic flow. We hence propose that this system, coupled with a restitution interface allowing localization in space (i.e. three-dimensional virtual sounds synthesis) can be used to restore essential visuomotor behaviors such as grasping desired objects and navigating (finding directions, avoiding obstacles) in unknown environments.