Abstract
Purpose:
Orienting and moving toward targets while perceiving trip hazards is important for safe and effective mobility. However, perceiving small, low-contrast, ground-based obstacles may be difficult in current prosthetic-vision devices using visual representations conveying only scene luminance (ie., intensity). We propose a new hybrid visual representation, Depth-cued Intensity (Int-D), that conveys the proximity of detected ground obstacles using a dedicated subset of phosphenes. The effectiveness of Int-D for orientation and mobility tasks was compared to standard intensity-based rendering (Int-S) in the presence of ground obstacles using simulated prosthetic vision with 8 noticeably different brightness levels. The simulated conditions model those recorded from a patient implanted with Bionic Vision Australia's 24 channel retinal prosthesis.
Methods:
8 normally-sighted (20/20, Pelli-Robson>=1.95) adults used a mobile artificial vision simulator with 17 phosphenes displayed centrally to both eyes and all external light blocked. For Int-D, the 5 bottom-of-view phosphenes conveyed the presence of obstacles in regions of the camera's lower visual field using depth to modulate phosphene brightness. Other phosphenes were rendered as per Int-S. Participants traversed a 4.5mx6.9m straight corridor (dark floor, white walls), containing gray ground obstacles varying in quantity, size, and placement, and a black wall target placed at eye-height as the destination point. The presentation order for all obstacle variables, start and target locations, and visual representations was randomised. Percentage of preferred walking speed (PWS), number of contacts, and distance from target were recorded.
Results:
Int-D (n=120, mean=0.208) had significantly fewer contacts than Int-S (n=120, mean=1.125, p<0.0001). No significant difference in speed was observed between Int-S (n=120; 29.81% of PWS; p<0.193) and Int-D (n=120; 33.32% of PWS). Int-D enabled participants to get significantly closer to the target (mean=0.529 metres) compared to Int-S (mean=0.932 metres, p<0.0001).The number of contacts increased significantly from low to high obstacle density placement for Int-S (p=0.044), but did not significantly change for Int-D.
Conclusions:
Depth-cued Intensity is robust to ground obstacles even at a high density and may be an effective strategy for enhancing orientation and mobility performance.
Keywords: 549 image processing •
584 low vision