Obesity treatment would benefit greatly from real-time monitoring of problematic eating behaviors in real-world settings. Wearable cameras provide significant utility in automatically detecting and understanding eating behaviors, yet pose major privacy concerns in real-world settings, significantly impacting longitudinal wear adherence. We objectively assess the wear time of three different video obfuscation methods in a crossover clinical trial to determine the greatest acceptability and feasibility, compared with no obfuscation. Lessons learned from this study have the potential to inform the design of privacy-conscious wearable devices that individuals with obesity will adhere to and that can accurately detect eating, which will advance our understanding of behavioral phenotypes surrounding obesity, laying the foundation for future timely interventions to reduce problematic eating behaviors.
30 participants wore a wearable camera for 2 free-living weeks with no explicit daily wear time requirement and reviewed clips of their video data each night. During 1 of the 2 weeks, participants’ clips were obfuscated by one of three randomly assigned methods (pixel blurring, pixel masking, and cartoon masking). The order of obfuscation and non-obfuscation weeks was counterbalanced across participants. The mean device wear times across each obfuscation method were compared to analyze the effect of each method on user adherence to the wearable camera. A one-way ANOVA was performed to evaluate the relationship between obfuscation methods and wear time. Poststudy interviews were conducted to assess user attitudes towards their assigned obfuscation methods.
Out of 30 participants, 17 completed the study. ANOVA was not significant between groups (ɑ=0.5, F=0.62, df=3, p=0.60). However, the cartoon masking method resulted in the highest mean wear-hours per week with 40.8 hours (SD=25.9), followed by pixel masking with 36.7 hours (SD=23.6), no obfuscation with 29.5 hours (SD=20.8), and, lastly, pixel blurring with 24.4 hours (SD=7.9). Cartoon obfuscation resulting in greatest adherence aligned with post-study interviews in which participants reported that superimposed avatars were extremely effective at obscuring sensitive and/or identifiable information. Participants did report that size and weight of the device impacted their wear time.
Our results suggest that cartoon masking may be more effective of the tested methods at assuaging privacy concerns regarding wearable cameras and thus increasing user adherence. Reasons for the lack of significance may be the low sample size, the weight and size of the device, and informing participants that they could take off the device any time. Future studies would benefit from using wear time as an objective method for assessing acceptability and should consider the lessons learned from WildCam.
We design, build, develop, and analyze. We are HABits Lab. HABits Lab is Affiliated With Northwestern Feinberg School of Medcine and Northwestern McCormick School of Engineering.