Abstract
Purpose: :
Perceiving trip hazards is critical for safe mobility and an important ability for prosthetic vision. Ground surface obstructions (e.g., bumps, steps, curbs) not marked by contrasting intensity or abrupt changes of depth can be difficult to perceive using standard prosthetic vision representations. These include representations conveying only the luminance (Intensity) or the relative depth of the environment (Depth), particularly with a low dynamic range visual representation. We have developed a novel system and method that estimates traversable space using binocular images to highlight trip hazards in a depth-based visual representation (Augmented Depth). We evaluate the effectiveness of augmented depth for mobility in the presence of ground obstacles using simulated prosthetic vision with 98 phosphenes and 8 noticeably different levels of brightness/size. The first generation retinal implant from Bionic Vision Australia will have 98 electrodes.
Methods: :
2 normally-sighted (20/20, Pelli-Robson>=1.95) participants used a mobile artificial vision simulator set with 98 phosphenes centrally displayed identically to both eyes to navigate a mobility course. A shroud blocked their normal vision. Phosphenes representing traversable ground were uniformly rendered with low brightness; all other phosphenes were rendered as per the standard depth-based representation, scaled up to increase contrast. Augmented Depth was compared to Depth and Intensity representations. Participants traversed a 9mx3m straight corridor (dark floor, white walls), containing black ground obstacles which varied in quantity, size, and placement. Trial obstacle variables and visual representation presentation order were randomised. Participants each completed 5x2 hour sessions consisting of approximately equal numbers of traversals for each visual representation (Augmented Depth: n=79; Depth: n=80; Intensity: n=79). The primary outcome measure was the number of contacts/collisions with obstacles and walls. Institutional Ethics Board approval was obtained.
Results: :
Augmented depth (mean = 0.346) performed significantly better than Depth (mean = 0.735, p=0.021) and Intensity (mean = 0.812, p = 0.007) with fewer number of contacts compared to Depth and Intensity visual representations.
Conclusions: :
Augmented Depth may be an effective representation for assisting mobility with current and future visual prostheses. Results show a significant advantage over intensity and depth-based representations for avoiding trip hazards.
Keywords: image processing • low vision