JMIR Publications

ABSTRACT

Background:

Unobtrusive observation is a promising method for assessing physical activity and other wellbeing behaviours in urban environments. However, current methods require the deployment of multiple in-person observers. Using video cameras instead could allow more accurate observations at a lower cost and with greater flexibility in observation scheduling.

Objective:

This research aimed to test the feasibility of using stationary wireless video cameras to observe physical activity and other wellbeing-related behaviours (e.g., social interactions), and assess its reliability and potential participant reactivity.

Methods:

Across three studies, a total of 148 hours of video recordings were collected from six outdoor public spaces in Manchester, UK. The videos were coded by three researchers using MOHAWk: a validated in-person observation tool for assessing physical activity, social interactions and people taking notice of the environment. Inter- and intra-rater reliabilities were analysed using intraclass correlation coefficients (ICCs). Intercept surveys were conducted to assess public awareness of the cameras and whether they altered their behaviour due to the presence of cameras.

Results:

The 148 hours of video recordings were coded in 85 hours. Inter-rater reliability was mostly excellent (ICCs > 0.90) between independent coders, although it was generally lower at night, remaining mostly excellent (ICCs > 0.90) or good (ICCs > 0.75). Intra-rater reliability within a single coder after a two-week interval was excellent (ICCs > 0.90) or good (ICCs > 0.75), indicating that the same coder could reproduce similar coding outputs over time. Intercept surveys with 86 public space users found no evidence of reactivity, suggesting participants did not alter their behaviour due to the presence of cameras.

Conclusions:

Camera-based observation methods are more reliable than in-person observation methods and do not produce participant reactivity typically associated with self-report. This method requires considerably less time for data collection and coding, and allows for safe observation at night without risk to research staff. This research is a crucial first step in demonstrating the potential for camera-based observation methods to improve and scale up natural experimental studies of real-world environmental interventions. It provides an important foundation for developing more scalable automated computer vision algorithms for assessing human behaviours.