Reproduce “Accessing Passersby Proxemic Signals through a Head-Worn Camera: Opportunities and Limitations for the Blind” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Kyungjun Lee, Daisuke Sato, Saki Asakawa, Chieko Asakawa, Hernisa Kacorri

Researchers from the University of Maryland, Carnegie Mellon University, and New York University empirically investigated the feasibility of providing blind individuals with real-time social interaction cues via head-worn cameras and computer vision. They found that while such systems offer valuable information, current implementations struggle with consistent visual capture and accurate estimation, particularly at close distances, highlighting a need for more robust computer vision and user-centric feedback mechanisms.

Key insights

  • Head-worn cameras worn by blind users face significant challenges in consistently capturing passersby, with head inclusion dropping from 98% beyond 6 feet to 56% within 6 feet due to user movements and camera limitations.
  • While computer vision successfully detected presence (1.0 precision) and relative position (0.87 F1-score), estimating distance (0.58 F1-score) and head pose (0.77 F1-score) proved less accurate, partly due to insufficient training data.
  • Participants found proxemic information valuable for social engagement and proactive decision-making, but highlighted issues with latency, verbosity, and the inability for blind users to verify or detect errors in system estimates.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.