Fig 1.
Workflow of the DeVAS software.
A: the original image of a step rendered by Radiance software. B: image filtered to simulate Severe low vision (VA 1.55 logMAR, CS 0.6 Pelli-Robson). C: luminance boundaries extracted from the filtered image. D: Pixels representing Geometric edges inferred from 3D data map of the space. E: Estimation of hazard visibility, based on a match between the luminance contours in C and the geometrical edges in D. Color coding represents the closeness of the match, ranging from red (poor match) to green (good match). F: A manually defined Region of Interest (ROI), G: The conjunction of E and F which specifies the hazard region of primary focus, which is used to generate the final Hazard Visibility Score (HVS).
Table 1.
Low-vision subject information.
Fig 2.
Geometry, lighting, and viewpoint variation of stimuli.
The top row used the lighting setting “spotlight 1” and viewpoint setting “center” to demonstrate the five target types: flat, big step-up big step-down, small step-up, and small step-down. The middle row used big step-down and center viewpoint to show the five lighting variations: overhead, far panel, near panel, spotlight 1, and spotlight 2. The bottom row used big step-down and spotlight 1 to show the five viewpoints: center, pivot left, pivot right, rotate down, and rotate up.
Fig 3.
Distribution of trials in ten HVS bins, each covering 0.1 width in a zero to one range.
The upper panel shows the trial distribution for Moderate blur (1.2 logMAR) and the lower panel shows the distribution for Severe blur (1.6 logMAR).
Fig 4.
Logistic regression model of aggregated data from subjects viewing with artificial blur.
Top: Moderate Blur (seven subjects), mean acuity 1.2 logMAR. Bottom: Severe Blur (seven subjects), mean acuity 1.6 logMAR. The red line shows the logistic regression function, transformed as shown in Eq 2. the gray area represents 95% confidence intervals.
Fig 5.
Histograms presenting correct and incorrect trials in each 0.1-wide bin of the HVS accumulated across seven subjects in each blur group.
The upper panel shows the distribution for moderate blur group trials, and the lower panel shows the distribution of severe blur group trials.
Fig 6.
Logistic regression models of 14 subjects with artificial acuity reduction by blur.
Top: Moderate Blur Group. Bottom: Severe Blur Group.
Table 2.
Individual regression models for subjects in moderate and severe blur groups.
Fig 7.
Logistic regression model of aggregated data from low-vision subjects.
The red line shows the regression curve and the gray area outlines the upper and lower bounds of the 95% confidence interval of the slope and intercept.
Fig 8.
Logistic regression models of 10 low-vision subjects.
Table 3.
Individual regression models of low-vision subjects.
Fig 9.
A scatterplot of logistic regression slope values of individual subjects and their visual acuities (upper panel) and contrast sensitivities (lower panel).
Linear regression trend lines are also plotted.
Fig 10.
ROI’s influence on visibility estimation.
A small downward step is shown in original resolution with three different definitions of ROI. From left to right, the first panel visualizes the visibility of the step with VA equivalent to 1.15 logMAR and CS 0.85 Pelli-Robson. Within this step, the left and right corners (green part) have high visibility, whereas the central horizontal edge (red part) has low visibility. The ROI can be defined as the side corners (second panel), or the central horizontal edge (third panel), or the whole step (fourth panel). The HVS derived from the corners ROI is 0.889, central ROI 0.051, while the ROI based on the entire step had HVS of 0.106. The definition of ROI will often lead to a significant change in HVS.