The purpose of this dissertation was to develop, validate, and implement multi-sensor approaches for measuring physical activity and social/contextual covariates in 2-5 year-old children via wearable-, wireless communication-, and infrared-depth camera-based technologies. In Chapter 2, a three-phased study design was used to validate a method for estimating metered distances between wearable devices using accelerometer-derived Bluetooth signals. Results showed that distances, up to 20 meters, can be predicted between a single Bluetooth beacon and receiver using a Random Forest algorithm. When multiple Bluetooth beacons and receivers were used within the same environment, a moving average filter was required to recover observations lost due to noise. Overall, simulation and validation data suggest that accelerometer-derived Bluetooth signals can be used in studies of physical activity co-participation to 1) estimate metered distances between devices using a single beacon-receiver paradigm, as well as to 2) estimate the proportion of time that devices are proximal when using multiple beacons and receivers. Chapter 3 characterized the relationship between objectively measured physical activity and dyadic spatial proximities in 2 year-olds and their parents. Data revealed that the overall proportions of time that children and their parents spent in total physical activity were positively associated, and time series data revealed that this relationship remained consistent when analyzed hour-to-hour. Time spent engaged in sedentary behavior was also positively associated between children and parents; however, there was no association between child and parent moderate-vigorous physical activity volumes. Dyadic proximity results showed that girls spent more time in joint physical activity with their mothers than boys. Furthermore, children who engaged in >60 minutes of daily moderate-vigorous physical activity spent an additional 30 minutes in joint total physical activity with their mothers each day, on average, when compared to children who engaged in <60 minutes of daily moderate-vigorous physical activity. Finally, boys and girls who engaged in >60 minutes of daily moderate-vigorous physical activity participated in joint physical activity with their mothers across wider relative distances, on average, than did boys who engaged in physical activity at closer relative distances to their mothers. In Chapter 4, an original computer vision algorithm was applied to infrared-depth camera data for the purpose of converting three-dimensional videos into triaxial physical activity signals in young children. Physical activity data were collected in 2-5 year-old children during 20-minute semi-structured, indoor child-parent dyadic play sessions. Play session video data were converted into triaxial physical activity signals using a multi-phased computer vision algorithm for each child. Computer vision-derived triaxial physical activity cut points for 2-5 year-olds were calibrated against a direct observation reference system using a machine learning algorithm. Results revealed that triaxial activity signals, as measured by a dual-sensor camera, can be used to estimate both physical activity intensities and volumes in young children without the use of wearable technology. Collectively, these studies show that multi-sensor approaches to physical activity measurement are a valid means by which to measure physical activity and social/contextual covariates in young children using either wearable sensors or computer vision.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/D8515F7B |
| Date | January 2018 |
| Creators | McCullough, Aston Kyle |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0019 seconds