Signal identification based on an eigenvector approach

In this paper, we propose a novel eigenvector-based signal identification algorithm for multi-dimensional signal identification. Signal patterns of 3-D accelerometer output concerning human activities are of low frequency, non-stationary and transient, and can also be termed dynamic or time-varying patterns of arbitrary length. Therefore, a matrix was formed by including features from each dimension of extracted signal pattern, and transformed eigenvectors associated with maximum eigenvalues were used as feature vectors in the identification process. Eigenvectors can preserve the identification efficiency of the feature matrix and can have the smallest number of features for robust, reliable classification in the application of multidimensional analysis.