Head-Mounted Mobility Aid for Low Vision Using Scene Classification Techniques

Abstract

This paper describes a new mobility aid for people with severe visual impairments which combines technology from the field of virtual reality with advanced computer vision techniques. A neural-network classifier is used to identify objects in images from a head mounted camera so that scene content specifically important for mobility may be made more visible. Enhanced images are displayed to the user on a head mounted display using a high saturation colour scheme where each type of object has a different colour, resulting in images which are highly visible and easy to interpret. The object classifier achieves a level of accuracy over 90%. Results from a pilot study conducted using people with a range of visual impairments are presented in which performance on a difficult mobility-related task was improved by over 100% using the system.