Wearable Camera High-Frequency Still Images Can Accurately Estimate Movement Behavior and Posture
Mentor 1
Scott Strath
Start Date
28-4-2023 12:00 AM
Description
Accurate and precise estimates of physical activity (PA) and sedentary behavior (SB) are important to fully elucidate the relationship between these behaviors and health. Currently available PA and SB measurement tools have limitations to use, and there is a scientific need to validate other objective PA and SB assessment measures. To examine the accuracy and precision of body-worn wearable camera (WC) still images to estimate PA/SB type and posture compared with direct observation. Each participant will be monitored twice in their own home for 1-hour at a time. Each monitored session is assigned a behavior domain to have the dataset represent a wide range of activities performed during daily living. Behavior domains are Household (cleaning, cooking), Active Leisure (exercising, sports), Inactive Leisure (watching television, using computer), and Community (shopping, running errands). During each monitored session participants will also be outfitted with the BRINNO WC, that will record still images in 1-sec increments. While the participant is carrying out different activity types within their chosen domain choice, a researcher will be video recording the participant using a Microsoft Surface Pro 6 with OBS Studio 28 software. The videoing component will consist of the researcher being approximately 6-ft away, and capturing all activities that the participant engages in. Each video and all WC still images will be annotated using a purposely designed annotation schema. The annotation schema will be detailing participant posture and activity type. The same codes will be used when annotating direct observation videos and still images from WC. Percent agreement and statistical bias will be calculated to determine accuracy and precision of WC still images. Confusion matrices will be conducted to determine where misclassifications occur.
Wearable Camera High-Frequency Still Images Can Accurately Estimate Movement Behavior and Posture
Accurate and precise estimates of physical activity (PA) and sedentary behavior (SB) are important to fully elucidate the relationship between these behaviors and health. Currently available PA and SB measurement tools have limitations to use, and there is a scientific need to validate other objective PA and SB assessment measures. To examine the accuracy and precision of body-worn wearable camera (WC) still images to estimate PA/SB type and posture compared with direct observation. Each participant will be monitored twice in their own home for 1-hour at a time. Each monitored session is assigned a behavior domain to have the dataset represent a wide range of activities performed during daily living. Behavior domains are Household (cleaning, cooking), Active Leisure (exercising, sports), Inactive Leisure (watching television, using computer), and Community (shopping, running errands). During each monitored session participants will also be outfitted with the BRINNO WC, that will record still images in 1-sec increments. While the participant is carrying out different activity types within their chosen domain choice, a researcher will be video recording the participant using a Microsoft Surface Pro 6 with OBS Studio 28 software. The videoing component will consist of the researcher being approximately 6-ft away, and capturing all activities that the participant engages in. Each video and all WC still images will be annotated using a purposely designed annotation schema. The annotation schema will be detailing participant posture and activity type. The same codes will be used when annotating direct observation videos and still images from WC. Percent agreement and statistical bias will be calculated to determine accuracy and precision of WC still images. Confusion matrices will be conducted to determine where misclassifications occur.
