Automated Person Detection in Dynamic Scenes to Assist People with Vision Impairments: An Initial Investigation

ASSETS 2018 |

Published by ACM

Related File | Related File

We propose a computer vision system that can automatically detect people in dynamic real-world scenes, enabling people with vision impairments to have more awareness of, and interactions with, other people in their surroundings. As an initial step, we investigate the feasibility of four camera systems that vary in their placement, field-of-view, and image distortion for: (i) capturing people generally; and (ii) detecting people via a specific person-pose estimator. Based on our findings, we discuss future opportunities and challenges for detecting people in dynamic scenes, and for communicating that information to visually impaired users,