Characterizing Walking Variability in the Real World using Wearable Sensors

Abstract

The walking movement of an individual is a window into their health. Both physical and mental illnesses can alter the way someone walks. As such, the analysis of walking gait is essential for understanding an individual's health status and mobility. Developments in measurement tools available in laboratory settings have enabled high-resolution, accurate, and comprehensive assessments of human movement. Researchers have established important relationships between key aspects of human walking and various health issues. However, the laboratory environment naturally places constraints on the types and variety of movements that can be measured; labs are not infinite in size, are indoors, and data collections are limited in time. As a result, researchers can only capture a small sample of the highly variable movement that people are capable of and exhibit when walking in the real world. This restricted view can result in decisions that aren't fully informed, as they are based on a mere fraction of available data. Thoroughly capturing the variability in movement patterns is essential to grasp the complete spectrum of an individual's abilities and performance. Doing so enables a more comprehensive understanding of health statuses and facilitates the delivery of better, more tailored care. The rapid advancement of wearable sensing technologies offers new possibilities for data collection outside the laboratory setting and in the real world. Wearable sensors of different types, sizes, and body placements exist and are accessible for the measurement of walking in unconstrained environments with little to no supervision. Real-world data is rich and can provide information that is not available in laboratory settings, such as habits or physical activity levels. However, walking in the real world is highly variable, involving a large variety of contexts. The movement patterns required to navigate the different scenarios of everyday life are diverse and can be complex. As such, interpreting the data collected in the real world using wearable sensors can be challenging. This dissertation proposes several frameworks that leverage real-world data collected from wearable sensors to address these challenges and further our understanding of how humans walk. Specifically, we collected datasets over multiple timescales and with different sensor suites to develop methods for analyzing walking variability in the real world. We identified various contextual factors that significantly impact human walking movement and energetics, and offer a new perspective on walking economy that complements previous findings established in the laboratory setting. Additionally, we demonstrate how a reduced sensor set, a single low-resolution accelerometer, can be utilized to characterize the variability of walking in the real world, by identifying parameters that can be used to parse the data in a meaningful way. We also built a framework able to accurately classify walking context using the same reduced sensor set. This work aims to facilitate and promote the use of wearable sensors for the study of human walking biomechanics in the real world. It demonstrates how capturing the variability in walking patterns can be an advantage, potentially providing insight into an individual's capabilities and health. The proposed methods have the potential to be extended to clinical populations and utilized by clinicians to aid their decision-making.