Wearable Indoor Personal Position And Activity Monitoring System

According to the data from the 6th national census, China has been in the rank of the Aging society. A decreasing birthrate coupled with a growing elderly population has led to an increased focus on elderly-centric issues such as long-term care and senior healthcare. However, the elderly care institutions are understaffed. It is unrealistic to provide a good and long-term care for the elderly with limited resources. In this case, wearable monitoring system enables real-time and long-term monitoring of indoor personal position and activities. Besides, the infrastructure-free wearable system only fixes small-sized sensors on human body. Consequently, the system is not only portable and acceptable by the elderly, but also cost-effective to setup and maintenance.This paper conducts research on indoor localization and activity recognition, respectively. Firstly, we realise a real-time activity recognition system based on only one Inertial Measurement Unit(IMU). The activities includes:stand, sit, lie, walk, run, sit-to-stand, stand-to-sit, lie-to-sit and sit-to-lie, among which the four later are transition movements. The improvement of activity recognition accuracy is made by the selection of fixed place of IMU and the features extracted from Euler angles. Secondly, we aim at increasing the external heading and position references for the classic inertial localization system framework. Based on the Zero-velocity Update (ZUPT) strategy, we formulate the acceleration error model and obtain the acceleration drift from the ac-cumulated velocity error which is reset at the beginning of the stance phase. Then the process of position estimation is rolled back to remove position error. Besides, IMU fixed on the waist are added to provide heading reference in the straight line walking and ranging sensors are fixed on two feet to provide position references for the other foot, respectively. Moreover, cooperative indoor localization algorithm is proposed in the case of many people moving around in the same building. Thirdly, the relationship of indoor position and activities is employed to formulate position-activity probability model and activity transition probability model, which fuse position and activity infor-mation for a higher accuracy of both localization and activity recognition. Activities recognized in this study include:make a phone call, drink water, cook, eat, teeth and so on. The experiments are conducted in the real indoor scenario, which is a mock elderly apartment. The results show a significant accuracy with position error less than 1 meter and activity recognition error at 15 %.