PHYSIOLOGICAL SIGNAL ANALYSIS FOR HOME-BASED ACTIVITY MONITORING
MetadataShow full item record
This dissertation develops a human-centric mobile framework to qualitatively monitor his/her physiological information and assess the quality of his activity. Human gait analysis is considered as the main domain of interest where the activity of the subject is monitored with a single wearable sensing unit that is an Inertial Measurement Unit (IMU). In addition to activity monitoring, this dissertation develops an output-only system identification method to further unravel the internal dynamics of such physical/physiological systems and their interactions with the environment. The dissertation consists of two main components. First, a robust sensor fusion method is proposed to accurately estimate the orientation and consequently position data to reconstruct the gait from a single IMU data. The proposed method of covariance inflated multiplicative extended Kalman filter (CI-MEKF) takes advantage of non-singularity of covariance in MEKF as well as a novel covariance inflation approach to fuse inconsistent information. The proposed covariance inflation approach compensates the undesired effect of magnetic distortion and body acceleration (as inherent biases of magnetometer and accelerometer sensors data, respectively) on the estimated attitude. The proposed CI-MEKF method is shown to be significantly robust against different uncertainties such as large body acceleration, magnetic distortion, and errors in the initial condition of the attitude and gyro bias. Based on the developed CI-MEKF filter, a template-based segmentation and classification technique is developed to monitor the daily activity of subjects. It extracts Mahalanobis distance-based features from multiple sections of the motion templates of each segmented data. Using a Support Vector Machine, the proposed wearable system can distinguish between nine classes of activities with a classification accuracy of 99.6%. It can also discriminate between normal and abnormal gait patterns with an accuracy of 98.7%. In addition to a high recognition rate, the proposed approach provides a Gait Similarity Score (GSS) of the performed gait to its desired/normal pattern. Second, a system identification approach is developed to identify the interaction dynamics of the physiological system such as musculoskeletal/nervous system in presence of unknown exogenous inputs. For instance, considering a human musculoskeleton during walking gait as a system and his/her muscle forces and interactions with the environment as unknown inputs, the dynamics of the system can be simplified into a multi-degree of freedom lumped model on which all active muscle forces and the ground reaction forces act as inputs. A similar problem statement is applicable to interactions between different regions of a nervous system and the unknown excitation as exogenous inputs that is perceived from the environment. In order to face such problems, an analytical output-only system identification (AOSID) method is developed, which is able to identify the coupled model of internal dynamics of a system and its exogenous inputs. Then, a few strategies are developed to decouple the model of internal dynamics of the system from its exogenous inputs. To facilitate the mathematical development, the model is assumed to be linear. To evaluate the AOSID performance, the method is applied to second-order MIMO (multi-input multi-output) systems. Simulation results on linear mechanical systems demonstrate that the AOSID is successful in identification and decoupling of the system from the inputs using only output data.