BubbleCam: Engaging Privacy in Remote Sighted Assistance

Document Title

BubbleCam: Engaging Privacy in Remote Sighted Assistance

Document Information

• Subject Area: Privacy protection and remote assistance system design
• Keywords: visually impaired individuals, privacy, remote visual assistance, computer vision, human-computer interaction

Research Background and Issues

• Problems and Challenges:

  • Remote Sighted Assistance (RSA) allows visually impaired individuals (PVI) to connect with volunteers in real time through cameras, but it may inadvertently expose private information, especially during live video interactions.
  • Existing solutions often focus on asynchronous image processing and lack exploration of privacy protection in real-time video contexts.
  • Technical barriers persist, such as the lack of accurate computational models for real-time private information recognition and the limited computational power of mobile devices.

• Significance:

  • Protecting the privacy of visually impaired individuals is a crucial step in enhancing their independence, confidence, and social participation.
  • There is a need to explore methods that not only protect privacy but also maintain usability to address both visual and privacy risks.

• Research Motivation and Related Work:

  • Although privacy protection technologies based on computer vision (e.g., blurring, face replacement) exist, they are primarily applied to image sharing rather than real-time video scenarios.
  • Multiple studies have confirmed the high demand for privacy management among visually impaired individuals, particularly regarding concerns about privacy breaches with temporary volunteers.
  • Given these circumstances, developing new real-time privacy protection tools is necessary to address the shortcomings of current RSA platforms.

Solution

• Proposed Method or Solution:

  • Designed a distance-based privacy protection tool, BubbleCam: using virtual bubbles (user-defined masking areas) to hide parts of the scene the user does not wish to share.

• Innovations:

  1. Introduced a simple and intuitive distance-based privacy protection mechanism without requiring users to manually label sensitive objects.
  2. Implemented real-time depth estimation and object masking using LiDAR technology, leveraging the ARKit support of mobile devices (e.g., iPad Pro).

• Implementation Steps and Technology:

  • The BubbleCam interface provides a slider that allows users to adjust the radius of the "virtual bubble."
  • The LiDAR scanner on iOS devices is used to quickly measure the distance between objects and the device. Combined with depth information, pixels beyond the threshold distance are masked or blurred.
  • Two hiding modes: Full Hidden mode and Partially Hidden mode.
  • The tool integrates with the iOS screen reader VoiceOver to meet the needs of visually impaired users.

Research Outcomes

• Specific Results:

  • Privacy Benefits:
    • 22 out of 24 participants expressed satisfaction with BubbleCam's privacy-enhancing features, noting that it significantly reduced unnecessary embarrassment while boosting confidence.
    • Users utilized it to mask personal items (e.g., ID cards, prescriptions) or cluttered room backgrounds.
    • Volunteers experienced reduced discomfort from viewing unnecessary or inappropriate content while assisting others.
  • Functionality and Usability:
    • BubbleCam successfully balanced privacy and assistance usability without hindering the core tasks of the interaction.
    • Users could adjust the masking range according to the task, ensuring the task focus remained visible.

• Comparison with Existing Solutions:

  • Compared to asynchronous privacy tools, BubbleCam offers the convenience of real-time operation, enhancing users' autonomy in privacy control.
  • It does not rely on complex models to label specific private objects, making it simple and broadly applicable.

• Experiments and Evaluation:

  • Findings were derived from a forward-looking field study involving 12 visually impaired users and 12 volunteers, with tasks including reading documents and selecting medications in common scenarios.
  • Analysis of users' bubble radius selection strategies revealed that low-vision users tended to choose smaller, more precise masking ranges compared to blind users.

• Limitations and Future Directions:

  • Limitations:
    • Limited duration of user testing, with certain features (e.g., switching hiding modes) being underutilized.
    • Experimental scenarios were confined to indoor simulations, requiring expansion to more real-world environments and different categories of private information, such as outdoor landmarks or personal attire.
  • Future research should explore more intelligent and personalized privacy controls, such as automated detection and selective masking based on user-defined privacy preferences.

Conclusion:
BubbleCam successfully protects the privacy of visually impaired individuals during remote assistance interactions, reducing the risk of information leaks while maintaining the usability of assistance services. It demonstrates the potential for a shift from unilateral privacy protection to collaborative privacy management, warranting further validation and expansion in broader contexts.