Distributed movement recognition algorithm based on wrist-mounted wireless sensor motes
Conference: European Wireless 2015 - 21th European Wireless Conference
05/20/2015 - 05/22/2015 at Budapest, Hungary
Proceedings: European Wireless 2015
Pages: 6Language: englishTyp: PDFPersonal VDE Members are entitled to a 10% discount on this title
Sarcevic, Peter; Kincses, Zoltan; Pletl, Szilveszter; Schaffer, Laszlo (Department of Technical Informatics, University of Szeged, Szeged, Hungary)
Movement data of the arm and body can be used in various applications, such as fall or emergency detection, or analysis of daily activity. Movement classification can be reliably done by using two wrist-mounted 9-degree-of-freedom (9DOF) sensor boards. These sensor boards are built up from a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, and are attached to wireless sensor network (WSN) motes. If the classification algorithm is implemented on only one of the motes, but uses the data from both sensor boards, data transfer via radio communication is required, which is very energy consuming. In this paper, a distributed algorithm is presented, where the motes use only their own data for classification, but the movement of the entire body and arms can be determined by combining the movement classes of the two arms. The proposed method requires smaller classifiers which can be easily implemented on low performance motes. Eleven movement classes were defined and data were collected in case of multiple subjects. The classes were divided by the role of the arm in the movement, and seven classes were defined for both arms. Various Time-Domain Features (TDF) were calculated in different processing window widths, and were combined in different configurations. Altogether 192 training and validation data sets were constructed for the two arms by different configurations of the sensors. Before the classification, dimension reduction was performed using the Linear Discriminant Analysis (LDA) method. The Minimum Distance (MD), and the MultiLayer Perceptron (MLP) classifiers were tested.