Comparison of different classifiers in movement recognition using WSN-based wrist-mounted sensors

Abstract

The analysis of human movement is a widely studied field of health applications, such as telerehabilitation, analysis of daily activities, and emergency detection. In this paper, the comparison of different classifiers is presented for a new movement recognition system, which can be used for the detection of emergency situations. The system uses 9-degree-of-freedom (9DOF) sensor boards that are attached to wrist-mounted Wireless Sensor Network (WSN) motes. The 9DOF sensor boards are built up from a tri-axial accelerometer, a tri-axial gyroscope, and a tri-axial magnetometer. Measurement data for classification were collected from multiple subjects. Eleven movement classes were constructed in order to recognize specific arm movements in stationary positions and also during the movement of the body. Various time-domain features (TDF) were calculated in different processing window widths. Depending on the used window size, sensors and TDFs, 48 different data sets were constructed, which were used for training and for validating of the system. Different classifiers were tested and compared using the original and the dimensionally reduced data sets. The dimension reduction is performed using the Linear Discriminant Analysis (LDA) method. The tested classifiers were the minimum distance classifier, the MultiLayer Perceptron (MLP) network, the naive Bayes classifier and the Support Vector Machine (SVM).