On-Site Assisted Photography
Marynel Vázquez & Aaron Steinfeld, 2009-2014
On-Site Assisted Photography

"A picture is worth a thousand words" when documenting problems encountered in the world ...


Research suggests photos are the preferred choice of transit riders for documenting problems in public transit. However, for people who are blind or low vision, taking a good picture can be problematic.

Methods for automatic image cropping, image adaptation for small displays, image or video retargeting are possible approaches for improving pictures. Nonetheless, these methods are designed for image post–processing, and tend to rely in composition heuristics that may not apply to photographers with visual impairments. For example, on-center compositions, where a dominant subject is geometrically centered in the image, are taken for granted in consumer photography and unlikely for users who are blind.


We propose a computational approach for assisting users with visual impairments during photographic documentation of transit problems.

The problem of taking a “good” picture is difficult, but dramatically simplified by the task characteristics. First, aesthetics are not an issue for problem documentation, thereby mitigating a significant challenge. Second, we do not need to know what the barrier is – we only need to know where it is. While being able to automatically annotate barriers might be useful for documentation, it is not essential. This mitigates the need for object recognition. Third, we can assume the rider is able to localize the barrier in space and roughly aim a camera at the target.


We integrate user interaction during the image capturing process, such that users can take better pictures in real time.

Assisted Photography Method

Our approach can be described as a method to avoid leaving out information that is expected to be most relevant. Motivated by Gestalt theories, we evaluate a meaningful group of contiguous salient points in a picture as potential region of interest (ROI). The system tries to center this region in a photo and correct excessive camera roll.



Sample code can be downloaded from GitHub.


This work was funded by grant number H133E080019 from the United States Department of Education through the National Institute on Disability and Rehabilitation Research.