Study On Human Activity Monitoring And Recognition By Using Body Sensor Networks

Body sensor network (BSN) is an extension of traditional wireless sensor network. BSN aims to provide ideal wireless setting for pervasive health care. Human activity monitoring is a major goal of BSN. By using on-body sensor, BSN could collect many kinds of signals from human movements. BSN could transmit the collected signals to a remote terminal. Existing researches have shown that human activity monitoring by using BSN could enable a range of health-related applications, including human activity recognition, fall detection, gait analysis, estimate of daily energy expenditure, and sports training.The main contributions of this study are described as follows:1. BSN usually uses many on-body sensors to monitoring human activities. In previous studies, the relationship between different sensor were not used for distinguishing human activ-ities. In this study, a recognition method based on coupled hidden Markov models (CHMMs) is proposed for multi-sensor data fusion to improve the recognition accuracy.2. Most previous recognition methods lack incremental learning ability. In this study, an incremental learning method is proposed. The proposed method is designed based on probabilis-tic neural networks (PNN) and an adjustable fuzzy clustering (AFC) algorithm. The proposed method establishes a new PNN from the initial training data, and update the PNN classifier from subsequent training data by using AFC to improve the recognition accuracy.3. Wearable accelerometers have played an important role in BSN. Installation errors of accelerometers may dramatically decrease the accuracy of recognition results. In this study, a method is proposed to calibrate acceleration signals. The method calculates a transformation ma-trix by using Gram-Schmidt orthonormalization to eliminate orientation errors, and then employs a low-pass filter to eliminate the main effect of the sensor's misplacement.4. Inertial sensors are widely used in BSN for human activity monitoring. In this study, an experiment is carried out to investigate whether inertial sensors could be used to measure abdominal muscles activity instead of traditional surface EMG. In this study, bridging exercises are performed. Radial basis function (RBF) networks were established to map inertial signals into EMG signals.