Julia Egly, “Comparison of Wearable Camera High Frequency Image Capture to Direct Observation Video Capture for Measuring Physical Activity Behavior”
Mentor: Scott Strath, Kinesiology
Poster #217

Direct observation (DO), a method of either observing a person or videoing a person, is considered the criterion standard to measure physical activity (PA) behavior. A limitation to using DO is that it is expensive and limited to small time observation windows. Examining alternative less expensive PA methods are able to assess PA behavior for longer windows of time is necessary. To determine if wearable camera (WC) high-frequency image capture can provide valid estimations of time spent in different PA behaviors compared to DO video. Twenty participants will complete two 1-hour home visits in which they will engage in their usual activities of daily living. During these activities, participants will be fitted with the Brinno TLC130 WC and video recorded with a Microsoft Surface Pro 8. WC images will consist of a first-person view, whereas the DO video will consist of a third person recording the participant view. WC high-frequency image capture will be converted to MP4 1 frame per sec video, to permit a first person versus third person comparison. Both WC and DO video will be annotated using an a priori developed coding schema derived from the Compendium of Physical Activities. Activities will be collapsed into broad behavioral domain categories for statistical analysis and annotations will be time matched. Overall percent agreement, confusion matrix, statistical bias, and a percentage agreement between WC video and DO video will be computed. Preliminary results from a 1-hour visit from 1 participant show the WC video agrees with DO video 94.1%. This visit included a range of behaviors such as exercise, laundry, meal preparation, and the use of electronics. WC first-person video capture has the utility to provide accurate estimates of PA behavior. Preliminary results across a variety of behaviors look promising. More participants will be recruited to calculate statistical bias.