On analyzing multiple, physiological sensor databases

Recent technological advances are encouraging a pervasive deployment of various physiological sensors such as motion trackers, accelerometers, EMG (Electromyogram), EKG (Electro-cardiograms), and other sensors that are used for monitoring and managing medical conditions as well as human performance in sports and military operations/training. The physiological data coming from multiple sensors are generally in time series format and forms multiple, multi-dimensional time series framework. Analyzing different physiological data in such integrated environment poses major challenges: different sensors have different characteristics, different people generate different patterns through these sensors, and even for the same person the data can vary widely depending on time and environment. In this dissertation, we assess the role of joint movements and muscular activities by proposing an approach to quantify integrated features of biomechanical kinematics with electrophysiology. We study content-based retrieval and cluster analysis of human motions based on the similarity of joint movements and electromyogram activity. Next, applications with multiple sensor systems require efficient access to large-scale, heterogeneous multi-dimensional data sets. To achieve this objective, we propose an integrated indexing structure based on bi-level spatial grids to efficiently support content-based queries on such multiple, high-dimensional data sets. Time series pattern mining (TSPM) finds correlations or dependencies in same series or in multiple time series. When the numerous instances of multiple time series data are associated with different quantitative attributes, they form a multiple multidimensional framework. Finding frequent patterns and ultimately, high confident association rules in such multidimensional environment is always a challenge. In this study, our emphasis is on discovering frequent patterns in multiple time series through sequential mining across varying time slices, and also mining quantitative attributes of only those time series that are present in the discovered patterns. Analyzing heterogeneous, high-dimensional physiologic and motoric streams to quantify the human performance and at the same time, provide visualization for performances of participants in low-dimensional space for easier interpretation is the another important issue discussed in this dissertation. We proposed an efficient, multidimensional factor analysis technique for analyzing and visualizing body sensor network data across different participants. Finally, in this study we discuss the experiments conducted on real-world data of human body motions and muscular functions that demonstrates the applicability of our approaches in practical applications such as physical medicines, medical rehabilitations, training and sports performances.