Retinal Prosthetics: Restoring Sight to the Blind Using Computer Vision
Over 285 million people around the world, almost the size of the entire US population, are blind or visually impaired. Retinal degenerative
diseases, which damage the photoreceptors in the retina, are one of the leading causes of visual impairment.
We have been developing retinal prosthetics to restore sight to those afflicted with retinal degenerative diseases. Retinal prosthetics
bypass the malfunctioning retinal photoreceptors, using a surgically implanting electrode array in the eye. Visual signals captured from a
camera are processed and then wirelessly sent to the surgically implanted electrode array in the eye. The electrode array then stimulates
the appropriate cells in the retina which pass the electrical signals down the visual pathway to the vision processing part of the brain.
However, while the retina has 125 million photoreceptors, conveying about 125 million pixels worth of visual information, the implanted
electrode array contains only 324 electrodes and thus can convey just 324 pixels of visual information. 324 pixels, or an 18x18 square, is
barely bigger than a 16x16 favicon found on browser tabs. This thesis focused on designing a suite of algorithms to convert 10-million-pixel
images from the blind user’s camera into 324-pixel images in real time, using limited computational power and passing on only the most
salient information. Ultimately, this thesis aims to help blind patients using retinal prosthetics walk around independently while avoiding
obstacles and recognize objects of particular interest in the scene. In this presentation, I will present previous work on retinal prosthetics
and demonstrate how our algorithms provide faster and more salient visual information to the blind to aid them in walking around