Validating a model of architectural hazard visibility with low-vision observers

Area of Science:

• Human-Computer Interaction
• Accessibility Research
• Computational Vision

Background:

• Pedestrians with low vision face increased injury risks due to low visibility of architectural hazards like steps and posts.
• Current methods for assessing hazard visibility in architectural designs lack computational precision.

Purpose of the Study:

• To validate a computational model designed to estimate the visibility of architectural hazards for individuals with low vision.
• To determine if the model's output scores can predict human observers' ability to identify hazards accurately.

Main Methods:

• The study involved two experiments using computer-generated architectural spaces and simulated low vision conditions.
• Experiment 1: 14 normally sighted subjects with simulated acuity reduction. Experiment 2: 10 low-vision subjects.
• Participants identified targets (steps, flat continuation) in 250 trials each, while the model generated visibility scores for comparison using logistic regression.

Main Results:

• A significant relationship was found between the model's visibility scores and the accuracy of hazard identification for 12/14 normally sighted subjects and all 10 low-vision subjects.
• The model's scores successfully predicted the likelihood of observers correctly identifying hazards under various lighting and viewpoint conditions.

Conclusions:

• The study provides strong evidence for the validity of the computational model in predicting architectural hazard visibility.
• This validated model serves as a foundational tool for architects to proactively assess and enhance the accessibility and safety of built environments for people with low vision.