On the use of the quaternion generalized Gaussian distribution for footstep detection

We propose a method to detect human footsteps from a vector-quaternion signal acquired by a tri-axial geophone. The quaternion generalized Gaussian distribution (QGGD) is derived to parameterize variations in the vector-quaternion signal using a shape parameter, quantifying non-Gaussianity and quaternion augmented covariance matrix, quantifying inter-channel correlation. The detection of footsteps is then formulated as binary hypotheses tests in terms of the parameters of the QGGD. The effectiveness of the proposed metrics is evaluated on recorded seismic data.