Wearable Cameras Reveal How Coding Rules Shape Children’s Screen Time Estimates

Screen time has become one of the most contested variables in child health research, blamed for everything from displaced physical activity to poor sleep and unhealthy eating. Yet the evidence behind those fears rests on a surprisingly shaky foundation: most studies still measure screen use with questionnaires, which depend on memory and are vulnerable to social desirability bias. A new study from the University of Otago in New Zealand, published in the Journal of Activity, Sedentary and Sleep Behaviors, tackles a deceptively technical question with major consequences for the field: when researchers strap a camera onto a child’s chest and let it snap photos every few seconds, how much do the choices made afterward, about how often images are captured and how obscured frames are handled, actually change the answer?

The research team, led by Rosie F. Jackson and Rachael W. Taylor, drew on data from the DREAM study, a randomized crossover trial that manipulated children’s sleep to see how mild sleep deprivation affected diet, activity, and wellbeing. Participants aged 8 to 12 wore a Brinno TLC130 camera on their chest, capturing a static image every two seconds from waking until bedtime over two days in each sleep condition. After applying strict criteria requiring matching data from both intervention conditions, 51 children, just over half of them girls and 14 percent indigenous Māori, provided usable data. The resulting dataset was enormous: 187 observations spanning more than 1.7 million images, covering time before school, after school, and on weekends, when recreational screen use is most likely to occur.

Every image was coded by hand using the free Timelapse 2 software, with two trained coders reaching better than 90 percent agreement before the full dataset was processed. Images were labeled for screens and screen indicators such as keyboards or remotes, for blocked or irrelevant views, and for whether a device was being actively used or merely sitting in the background. Crucially, the researchers made a methodological decision that sets their work apart: instead of baking rules about ambiguous images into the coding protocol itself, they coded each image individually and applied all decision rules afterward, during data processing. That separation meant they could test different assumptions against the same raw data without recoding anything, a luxury most wearable camera studies have never had.

The first question the team examined was how the interval between images affects screen time estimates. Because the cameras had captured a frame every two seconds, the researchers could simulate sparser sampling by systematically discarding images, recreating what a study would have recorded at intervals of 4, 6, 8, 10, 20, 30, or 60 seconds. The results were striking. At the group level, estimates of average screen time barely moved, no matter how widely the frames were spaced. Median differences compared with the dense two-second data were all under two minutes, meaning a study that sampled once a minute instead of twice a second would still arrive at essentially the same population average.

Individual accuracy told a different story. When the researchers looked at how far each child’s estimate deviated from the two-second benchmark, longer intervals introduced more scatter. A 10-second interval emerged as the sweet spot: half of the children had estimates within a minute of the dense-data value, while the workload dropped dramatically. Moving from two-second to 60-second intervals cut the number of images requiring coding to just 3 percent of the original total, a reduction that could shrink months of manual annotation into days. For studies that need reliable data on individual children, such as randomized crossover designs, the message is clear: sample at 10 seconds. For studies that only need group averages, once a minute is enough.

The second question concerned blocked images, one of the messiest realities of wearable camera research. Children turn sideways to talk, slump at the table so the camera points at the floor, or bury themselves under blankets while watching television, leaving the lens staring at fabric or the ceiling. In this dataset, between 4.6 and 19 percent of valid images were blocked. The team tested two rules for handling these gaps. Rule 1, a conservative approach, reclassified up to 10 consecutive no-screen images, a maximum of 20 seconds, as screen time, provided the same device and activity appeared on both sides of the gap. Rule 2 was a deliberate worst-case scenario, allowing unlimited consecutive blocked images to be counted as screen time under the same condition.

The impact of these rules was substantial. Applying Rule 1 increased weekend estimates of total screen use by a median of 8.8 minutes, roughly a 5 percent bump, while the unlimited Rule 2 added a median of 56.4 minutes, an increase of around 32 percent. Effects were smaller before school, when time pressure naturally limits screen opportunities, and larger during after-school and weekend hours when children had freedom to settle in for long sessions. The authors note that Rule 2 likely overestimates true use, since some blocked episodes genuinely involve no screens, but the exercise demonstrates just how much a seemingly innocuous coding assumption can inflate or deflate headline figures. Previous studies have used tolerances ranging from two to 4.5 minutes for blocked images, and because those rules were embedded in coding schedules, their true effect on estimates has never been quantifiable.

The study also distinguished between total screen time, which includes devices visible in the background, and priority screen time, where the child is actively engaged. Priority estimates ran about 10 percent lower than total estimates, a distinction that matters when comparing objective data with questionnaire responses. Parents and children are unlikely to report a television murmuring in the corner of the room, so classifying engagement levels allows researchers to align wearable camera data with the subjective measures that dominate the existing literature, and to understand exactly where the two approaches diverge.

The implications reach beyond screen time research. Wearable cameras are increasingly used to study food marketing exposure, sedentary behavior, and physical activity, and all of these fields face the same unresolved questions about sampling density and missing data. This study provides the first direct evidence that longer capture intervals are far less damaging than commonly assumed, at least for group-level estimates, and that processing rules for obscured images can shift results by tens of minutes per day. The authors recommend that future researchers capture images at 10-second intervals when individual-level accuracy matters, tolerate intervals up to a minute for group comparisons, and, critically, apply assumptions about blocked images during data processing rather than during coding, preserving transparency and the ability to test alternatives.

Challenges remain. Manual coding of millions of images carries a heavy researcher burden, limiting the method to smaller studies where accurate measurement justifies the cost, and the authors point to machine learning as the eventual path forward, though privacy and reliability concerns still need solving. The strict crossover requirements also halved the eligible sample, and the included children differed somewhat from the full cohort in maternal education and household deprivation, though the authors argue this is unlikely to affect conclusions about processing rules. Ethical safeguards, including participant review of images before researchers saw them and secure storage, were followed throughout. What the study delivers is a rare thing in measurement science: a quantified map of how methodological choices ripple through the final numbers. As debates over children’s screen use continue to shape policy and parenting advice, this work suggests the field can now measure the phenomenon with far more confidence, and far fewer photographs, than anyone assumed.

Subject of Research: Objective measurement of children's screen time using wearable cameras and the effect of image capture intervals and data processing rules

Article Title: The influence of different processing rules on wearable camera data estimates of habitual screen time in children

Article References: Jackson, R. F., Meredith-Jones, K. A., Haszard, J. J., Galland, B. C., Morrison, S., Jaques, M., & Taylor, R. W. (2026). The influence of different processing rules on wearable camera data estimates of habitual screen time in children. Journal of Activity, Sedentary and Sleep Behaviors, 5(1), Article 3. https://doi.org/10.1186/s44167-026-00095-1

Image Credits: AI Generated

DOI: 10.1186/s44167-026-00095-1

Keywords: screen time, wearable cameras, children, objective measurement, data processing, image coding, sedentary behavior, sleep deprivation, methodology, digital devices, child health, New Zealand