Energy-Efficient Posture Classification with Filtered Sensed Data from A Single 3-Axis Accelerometer Deployed in WSN

Laurentiu Hinoveanu, Jacek Lewandowski, Xiang Fei, Hisbel Arochena, Partheepan Kandaswamy, Zhipeng Dai
unpublished
Wireless Sensor Networks (WSNs) provide rich and detailed measurements of the physical phenomenon that they monitor. With the sensed data, a variety of pervasive applications can be developed. One category of those applications are classifications based on supervised machine learning, with one example being postures recognition with data from body sensor networks (BSNs). Conventionally for accuracy reason, raw data from the BSN sensors, such as accelerometers or other inertial devices, is
more » ... itted to the central unit for postures identification. It has been well known, however, that in most of the battery powered WSNs, communication consumes most of the energy. This paper explores the possibility of obtaining the same level of classification accuracy with the filtered sensed data to prolong the lifetime of the WSNs. A special case of posture recognition based on Artificial Neural Networks (ANN), Naive Bayes and K-Nearest-Neighbours (KNN) has been studied, and a mechanism for the posture classification based on filtered sensed data has been constructed. Real data from a Shimmer node has been collected and the test results show that the same level of accuracy can be obtained using only two thirds of the raw data. The implementation considerations with some prototypes have also been provided.
fatcat:mrz6vb5pzffy7o3wulok3rwotu