Search TESS:

.

Can masking of video data reduce perceived invasion of privacy?


Download data and study materials

Download proposal

 


Principal Investigator(s):

Julie Downs
Carnegie Mellon University
Email: downs@andrew.cmu.edu
Home page: http://www.cmu.edu/dietrich/sds/people/faculty/julie-downs.html

Sample size: 1800
Field period: 5/25/2010-10/5/2011

 

Abstract:

Robotic systems using vision-driven technologies offer solutions to societal and individual problems, but bring real concerns about the potential for invasion of privacy. But robotic systems rely on only a fraction of the information found in full-motion video. Thus, there may be traction to be gained by reducing the information gathered in video format, omitting extraneous data that would capture private information but be irrelevant to the robotic system. Such approaches only make sense if people’s sense of privacy invasion is truly lessened by limiting the data. Data obscuration is used to protect privacy in large datasets containing sensitive information or personal identities. With full-motion video, large segments of the population are resistant to considering its use out of concerns about their privacy. To date, there has been very little investigation into how people perceive different visual representations of their bodies and their actions, even as designers attempt to incorporate privacy in the initial development of their systems. The current research aims to investigate whether different approaches to visual masking of captured video result in mitigated concern for privacy, both as a test of the general concept and as a preliminary step in identifying parameters most associated with privacy concerns in visual technologies.

Hypotheses:

1. Masking techniques applied to video data will reduce perceptions of invasion of privacy associated with video monitoring systems

2. Reductions in perceptions of invasion of privacy will increase acceptability of video monitoring systems

Experimental Manipulations:

Presentation of video monitoring system with different degrees of masking:
1- No masking (full-motion video, used to assess floor-level tolerance of acceptability of a monitoring system using video)
2- Blurred video
3- Stick figure information extraction video
4- Hypothetical monitoring without video (black screen, used to assess ceiling-level tolerance of acceptability of any monitoring system)

Key Dependent Variables:

1. Perceptions of invasiveness (3 questions on 7-point scales)
2. Acceptability of system (1 question on a 7-point scale)
3. Whether the respondent would use such a system (1 yes/no question)

Summary of Findings:

Using a between-subjects design, we included conditions approximating floor and ceiling values in order to define the space in which video-obscuring (masking) techniques could operate. Past research using within-subjects methods had estimated this distance to be substantial, although these estimates were likely exaggerated by contrast effects inherent in such a design. Our data indicate that the distance between floor and ceiling, when presented singly, is actually quite narrow, ranging from 71% using the system when presented as floor to 79% at ceiling (p=.003, with similarly narrow differences in ratings of acceptability, p=.018, and invasiveness, p<.001). A final, within-subjects set of questions comparing all conditions did replicate the much larger effect size, with the gap between those using the system presented as floor versus ceiling increasing from 8% to over 40%.


Within this narrow range, the stick figure slightly outperformed the floor level (full-motion video) in ratings of invasiveness (p=.001), but only marginally in the proportion saying that they would use the system (76%, p=.082). There were no differences in how acceptable the stick figure video was seen compared to floor level (p>.10). The stick figure video was not significantly worse than the ceiling (no video) condition on any measures (all p>.10).


Surprisingly, the video condition where we obscured information by blurring the video was perceived as worse on all measures compared to the floor condition. It was perceived as marginally more invasive (p=.087), less acceptable (p=.066) and less likely to be used (p=.077) than full-motion video. Although none of these differences quite reached statistical significance, the consistency of the unexpected pattern is informative.


Ratings of acceptability and an index score of invasiveness both significantly predicted uptake of the system (both p<.001), relationships that were unaffected by controlling for age and whether the respondent was a caregiver for an adult dependent.
These results suggest that methods for obscuring video content can be useful in alleviating participants concerns about invasiveness, although the effects on acceptability and uptake of such a system are less robust. Furthermore, different methods of masking content may have extremely different effects, with some backfiring rather than alleviating concerns.


Copyright © 2014, TESS