Classification of Human Movement based on Radio Signals in Wireless Body Area Network using Artificial Neural Networks

Abstract

This paper aims to classify the signals acquired using wireless sensors placed on Human body for various human activities such as running, walking and standing. The proposed method can identify the three human movements using the radio signals transmitted using wireless sensors. Open NICTA provides the BAN (Body Area Network) measurement channel in three kinds of human motions. This paper used all nine sets of the measurement data for each human activity with respect to transmitter-receiver from back chest, right wrist - chest, right ankle - chest, chest - right hip, right wrist  - right hip, left wrist - right hip, right ankle - right hip, left ankle - right hip and back - right hip. All the dataset was separately processed for individual human activity and then the features were combined for training the classifier. The feature set comprised of 11 components along the second dimension of each signal with a window of 25 samples thus reducing the samples to 160 from 4000. Only 27 front line features were considered form 99 to reduce the dimensionality while securing accuracy. The selection of only 27 features was settled after inspection about the deviation of each sample from the mean. The system achieved an accuracy of almost 99% using Artificial Neural Networks for the test samples which was 25% of the total dataset. The cross validation showed an average accuracy of 99%.