Article
Detection of human faces by blind patients implanted with the Argus® II Retinal Prosthesis System
Search Medline for
Authors
-
Stefania Guerra
- Second Sight Medical Products, Sàrl, Lausanne, Switzerland
-
P. Stanga
- Manchester Vision Regeneration (MVR) Lab, Manchester Royal Eye Hospital, Manchester, UK; Manchester Academic Health Science Centre and Centre for Ophthalmology and Vision Research, Institute of Human Development, University of Manchester, Manchester, UK
-
F. Merlini
- Second Sight Medical Products, Sàrl, Lausanne, Switzerland
-
J. Sahel
- CHNO des Quinze-Vingts, INSERM-DHOS CIC, Paris, France; Institut de la Vision, CNRS, UMR_7210, Paris, France
-
S. Mohand-Said
- CHNO des Quinze-Vingts, INSERM-DHOS CIC, Paris, France; Institut de la Vision, CNRS, UMR_7210, Paris, France
-
L. da Cruz
- Moorfields Eye Hospital, Moorfields Eye Hospital, London, UK
-
A. Caspi
- Second Sight Medical Products, Inc, Sylmar, USA
-
R. Greenberg
- Second Sight Medical Products, Inc, Sylmar, USA
| Published: | February 13, 2014 |
|---|
Outline
Text
Purpose: To investigate whether Argus II subjects can locate human faces with their systems using a facial detection algorithm and whether detection speed improves when field of view that is mapped onto the Argus II implant is changed (i.e. demagnified).
Methods: To date, more than 50 patients blinded by outer retinal dystrophies received an Argus II epi-retinal prosthesis (Second Sight, Sylmar, CA). In normal use, a micro camera mounted on a pair of glasses gathers visual information. The video is subsampled to match the field of view of the implanted array and processed into 60 pixels that characterize the average brightness of the scene at each electrode location. In the current study, the image of the scene acquired by the video camera was processed using a face detection algorithm, resulting in a visual percept only where a human face was detected by the processor. A printed image of a face at normal size was place at random location on a wall at a distance of 3 meters. A distractor image with equivalent size and brightness was also placed on the wall at the same height. The subject was required to search for the face. In some trials, the image processing algorithm captured a field of- view that matched the field-of-view of the implanted array (20 degrees diagonally) while in some trials the entire field-of-view of the camera (53 degrees) was captured and “zoomed out” to fit the array. In a second experiment the blind subject was engaged in a conversation with a sighted person, who either faced the subject or turned away at some point during the conversation. The blind subject reported whenever he was unable to detect the location of the face.
Results: The patients tested were able to find the face with both magnifications. The time to find the target was significantly shorter when using the wider field-of-view. In the “real conversation” task, the blind subjects were able to recognize within a few seconds when the other person turned away.
Conclusions: Face detection in real world, i.e. at 2-3 m distance is a challenging task with a retinal implant. Using a device that takes advantage of external image processing, we can provide face detection functionality to blind patients. This feasibility study demonstrated that image processing algorithms can enable patients to perform daily tasks that are not limited by the resolution or the sensitivity of the array.
