What Makes Videos Accessible to Blind and Visually Impaired People?

Document Title

What Makes Videos Accessible to Blind and Visually Impaired People?

Document Information

• Topic Area: Research on video accessibility, specifically improving the video usage experience for blind and visually impaired (BVI) individuals
• Keywords: blind, visual impairment, online video, accessibility, video search, video description, automated scoring

Research Background and Problems

• What problems or challenges did the authors identify?

  • Online videos, as a primary source of information, are largely inaccessible to blind and visually impaired (BVI) individuals due to the lack of additional descriptions for visual content.
  • BVI users often rely on trial-and-error methods to preview videos, which consumes significant time. Current video platforms do not clearly indicate which videos are suitable for BVI users.
  • There is no automated method to quantify video accessibility.

• Why is this issue important?

  • The widespread use of online videos has made them a core medium for communication and information acquisition. However, BVI individuals face challenges in effectively selecting or understanding these contents. This barrier impacts their ability to access information and limits their participation in social interactions and entertainment.

• Research Motivation and Related Work

  • Existing research primarily focuses on providing manually or automatically generated audio descriptions, but these approaches have limited coverage and scalability.
  • The authors aim to develop an automated accessibility assessment tool that calculates accessibility scores based on speech, visual, and audiovisual interaction elements, helping BVI users quickly find suitable videos.

Solutions

• What methods or solutions did the authors propose?

  • The authors proposed seven accessibility heuristic principles and translated them into seven automated quantitative metrics:
    • Audio-related metrics:
      • Percentage of non-speech duration (% Non-speech)
      • Percentage of low lexical density speech (% Low lexical density speech)
    • Visual-related metrics:
      • Rate of shot changes
      • Number of visual entities per minute (# Visual entities / min)
    • Audiovisual interaction metrics:
      • Percentage of visual entities not described in speech (% Visual entities not in speech)
      • Number of undescribed on-screen text per minute (# Undescribed on-screen text / min)
      • Number of visual references per minute (# Visual references / min)
  • Developed an enhanced video search interface that displays video accessibility scores and supports filtering, optimizing the search experience for BVI users.

• What are the innovative aspects of this solution?

  • Integration of data-driven automated accessibility metrics to quantify video accessibility for BVI users.
  • Provision of simple, interpretable accessibility scores and metric explanations to support informed decision-making by BVI users.
  • Incorporation of accessibility predictions into the video search interface, reducing the time spent on trial-and-error methods.

• What are the implementation steps and key technologies used?

  1. Conducted interviews with BVI individuals to collect features relevant to accessibility assessment.
  2. Designed corresponding automated metrics for these features (e.g., speech transcription, visual content recognition, optical character recognition).
  3. Used linear regression to analyze the relationship between video accessibility scores and the seven metrics.
  4. Designed an enhanced video search interface and validated its effectiveness through user studies.

Research Outcomes

• What specific outcomes were achieved?

  1. Summarized seven accessibility heuristic principles and proposed corresponding quantitative metrics.
  2. Established a regression model with an adjusted ( R^2 = 0.642 ), indicating the model effectively explains BVI users' perceived accessibility.
  3. Developed a video search interface supporting accessibility scoring and filtering, significantly reducing the time and number of previews required to find suitable videos.

• How does it compare to existing solutions?

  • Automated metrics and scoring reduce the subjective workload of evaluating video accessibility.
  • Accessibility predictions and presentation improve search efficiency and user experience.
  • Complements existing audio description methods by addressing broader aspects of visual content accessibility.

• What were the experimental or evaluation results?

  • The enhanced video search interface significantly reduced search time (by 40%) and trial-and-error attempts (by 54%).
  • Users rated the prediction results as accurate (mean absolute error of 0.53) and strongly preferred the added accessibility features.

• Limitations and Future Directions

  • Limitations:
    • The dataset is relatively small (55 videos) and primarily sourced from YouTube's trending page, which may bias results toward more accessible videos.
    • Some metrics (e.g., visual entity count) did not directly impact perceived accessibility, potentially underestimating the importance of visual factors.
  • Future Work:
    • Expand the video dataset to include a wider range of topics and production types.
    • Explore the gap between perceived accessibility and actual accessibility.
    • Develop more efficient automated processing tools, such as segment-level video analysis.
    • Extend support to video platforms beyond YouTube, including social media platforms like Twitter and Facebook.